plateform
stringclasses 1
value | repo_name
stringlengths 13
113
| name
stringlengths 3
74
| ext
stringclasses 1
value | path
stringlengths 12
229
| size
int64 23
843k
| source_encoding
stringclasses 9
values | md5
stringlengths 32
32
| text
stringlengths 23
843k
|
|---|---|---|---|---|---|---|---|---|
github
|
domingomery/Balu-master
|
Bio_manyplot.m
|
.m
|
Balu-master/InputOutput/Bio_manyplot.m
| 1,215 |
utf_8
|
f88fda7a9bd7b5d2f13dd5d3ad52e3d2
|
% Bio_manyplot(x,y,labels,xlab,showper)
%
% Toolbox: Balu
%
% Plot of many (x,y) graphs
%
% Input:
% - x could be a column vector with n elements or a matrix with
% nxm elements
% - y is a matrix with nxm elements
% Output:
% - a plot of m curves, each curve has n points with different colors
%
%
% Example 1: same x for different y
% clf
% x = (0:0.01:1)';
% y = [sin(x*2*pi) cos(x*pi) sin(x*pi)];
% Bio_manyplot(x,y)
% grid on
% legend({'curve-1','curve-2','curve-3'})
%
% Example 2: different x for different y
% clf
% x = [sort(rand(100,1)) sort(rand(100,1)) sort(rand(100,1))];
% y = [sin(x(:,1)*2*pi) cos(x(:,2)*pi) sin(x(:,3)*pi)];
% Bio_manyplot(x,y)
% grid on
% legend({'curve-1','curve-2','curve-3'})
%
% (c) Domingo Mery, 2019
% http://dmery.ing.puc.cl
function Bio_manyplot(x,y)
hold on
set(gca,'FontSize',12)
sc = 'bgrcmykbgrcmykbkbgrcmy';
sq = 'sdvo^>sdvo^>vo^>sd';
sl = '-:-:-:-:-:-:-:-:-:-:-:-:-:';
p = size(y,2);
m = size(x,2);
if m==1
x = repmat(x,[1 p]);
end
for i=1:p
plot(x(:,i),y(:,i),[sq(i) sl(i)],'LineWidth',2,...
'MarkerEdgeColor',sc(i),...
'MarkerFaceColor',sc(i+1),...
'MarkerSize',5)
end
|
github
|
domingomery/Balu-master
|
Bio_plotfeatures.m
|
.m
|
Balu-master/InputOutput/Bio_plotfeatures.m
| 4,944 |
utf_8
|
ef2f13c1a1d21d79d578bfa400278972
|
% Bio_plotfeatures(X,d,Xn)
%
% Toolbox: Balu
% Plot features X acording classification d. If the feature names are
% given in Xn then they will labeled in each axis.
%
% For only one feature, histograms are ploted.
% For two (or three) features, plots in 2D (or 3D) are given.
% For m>3 features, m x m 2D plots are given (feature i vs. feature j)
%
% Example 1: 1D & 2D
% load datagauss % simulated data (2 classes, 2 features)
% figure(1)
% Bio_plotfeatures(X(:,1),d,'x1') % histogram of feature 1
% figure(2)
% Bio_plotfeatures(X,d) % plot feature space in 2D (2 features)
%
% Example 2: 3D
% load datareal % real data
% X = f(:,[221 175 235]); % only three features are choosen
% Bio_plotfeatures(X,d) % plot feature space in 3D (3 features)
%
% Example 3: 5D (using feature selection)
% load datareal % real data
% op.m = 5; % 5 features will be selected
% op.s = 0.75; % only 75% of sample will be used
% op.show = 0; % display results
% op.b.name = 'fisher'; % definition SFS with Fisher
% s = Bfs_balu(f,d,op); % feature selection
% Bio_plotfeatures(f(:,s),d) % plot feature space for 5 features
%
% See also Bev_roc.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function Bio_plotfeatures(X,d,Xn)
if ~exist('Xn','var')
Xn = [];
end
m = size(X,2);
if m>9
error('Bio_plotfeatures for %d features makes %d plots.',m,m*m)
end
scflag = 0;
if size(d,2)==2
sc = d(:,2);
d = d(:,1);
scflag = 1;
end
dmin = min(d);
dmax = max(d);
col = 'gbrcmykbgrcmykbgrcmykbgrcmykgbrcmykbgrcmykbgrcmykbgrcmykgbrcmykbgrcmykbgrcmykbgrcmyk';
mar = 'ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^ox+v^';
% clf
warning off
r = sprintf('%s',39);
warning on
s = 'legend(';
for k = dmin:dmax
s = [s r sprintf('class %d',k) r ];
% s = [s r sprintf('class %d',k-1) r ];
if k<dmax
s = [s ','];
else
s = [s ');'];
end
end
if (m<4)
switch m
case 1
for k = dmin:dmax
[h,x] = hist(X(d==k),100);
dx = x(2)-x(1);
A = sum(h)*dx;
x = [x(1)-dx x x(end)+dx];
h = [0 h 0];
plot(x,h/A,col(k+1));
hold on
end
if ~isempty(Xn)
xlabel(Xn);
else
xlabel('feature value');
end
ylabel('PDF');
case 2
if isempty(Xn)
Xn = ['feature value 1';'feature value 2'];
end
for k = dmin:dmax
ii = find(d==k);
plot(X(ii,1),X(ii,2),[col(k+1) mar(k+1)]);
hold on
if scflag
for ic=1:length(ii)
text(X(ii(ic),1),X(ii(ic),2),[' ' num2str(sc(ii(ic)))]);
end
end
end
title('feature space');
xlabel(Xn(1,:));
ylabel(Xn(2,:));
case 3
for k = dmin:dmax
ii = find(d==k);
plot3(X(ii,1),X(ii,2),X(ii,3),[col(k+1) mar(k+1)]);
hold on
if scflag
for ic=1:length(ii)
text(X(ii(ic),1),X(ii(ic),2),X(ii(ic),3),[' ' num2str(sc(ii(ic)))]);
end
end
end
if ~isempty(Xn)
xlabel(Xn(1,:));
ylabel(Xn(2,:));
zlabel(Xn(3,:));
else
xlabel('feature value 1');
ylabel('feature value 2');
zlabel('feature value 3');
end
end
eval(s)
else
l = 1;
for j=1:m
for i=1:m
zi = X(:,i);
zj = X(:,j);
subplot(m,m,l); l = l+1;
for k = dmin:dmax
ii = find(d==k);
plot(zi(ii),zj(ii),[col(k+1) mar(k+1)]);
hold on
if scflag
for ic=1:length(ii)
text(zi(ii(ic)),zj(ii(ic)),[' ' num2str(sc(ii(ic)))]);
end
end
end
if ~isempty(Xn)
xl = Xn(i,:);
yl = Xn(j,:);
else
xl = sprintf('z_%d',i);
yl = sprintf('z_%d',j);
end
if i==1
ylabel(yl)
end
if j==m
xlabel(xl)
end
end
end
end
|
github
|
domingomery/Balu-master
|
Bio_sendmail.m
|
.m
|
Balu-master/InputOutput/Bio_sendmail.m
| 1,611 |
utf_8
|
55f1122029d857abda5f30fe1ef3f6de
|
% Bio_sendmail(mymail,mypassword,mailto,subject)
% Bio_sendmail(mymail,mypassword,mailto,subject,message)
% Bio_sendmail(mymail,mypassword,mailto,subject,message,attachment)
%
% Toolbox: Balu
% Send e-mail
%
% Send an e-mail form mymail to mailto, with subject, message (optionsl)
% and attachment (optional). Ir requires the password of mymail.
%
% Example 1:
%
% x=10;y=20;
% msg1 = sprintf('x=%d',x)
% msg2 = sprintf('y=%d',y)
% Bio_sendmail('[email protected]','johnssszzz12','[email protected]','New results',{'Hi Mary,','here are the results:',msg1,msg2,'John'});
%
%
% Example 2:
%
% Bio_sendmail('[email protected]','johnssszzz12','[email protected]','Output Image','Hi Mary, I am sending you the image. John','image.jpg');
%
% See also sendmail.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function Bio_sendmail(mymail,mypassword,mailto,subject,message,attachment)
fprintf('Sending e-mail to %s...\n',mailto);
setpref('Internet','E_mail',mymail);
setpref('Internet','SMTP_Server','smtp.gmail.com');
setpref('Internet','SMTP_Username',mymail);
setpref('Internet','SMTP_Password',mypassword);
props = java.lang.System.getProperties;
props.setProperty('mail.smtp.auth','true');
props.setProperty('mail.smtp.socketFactory.class','javax.net.ssl.SSLSocketFactory');
props.setProperty('mail.smtp.socketFactory.port','465');
if exist('message','var')
if exist('attachment','var')
sendmail(mailto,subject,message,attachment);
fprintf('>> attaching file %s...\n',attachment);
else
sendmail(mailto,subject,message);
end
else
sendmail(mailto,subject);
end
|
github
|
domingomery/Balu-master
|
Bio_findex.m
|
.m
|
Balu-master/InputOutput/Bio_findex.m
| 436 |
utf_8
|
ce9b6a520e7a8e6ea15dd58439740828
|
% Obtain the index number of the feature wich names contain a given string.
function [ix,fnix] = Bio_findex(fn,str,inc)
if ~exist('inc','var')
inc = 1;
end
[N,M] = size(fn);
n = length(str);
T=ones(N,1)*str;
ix = [];
for i=1:M-n+1
D = sum(abs(T-fn(:,i:i+n-1))')';
ii = find(D==0);
if ~isempty(ii)
ix = [ix;ii];
end
end
if not(inc)
ii = (1:N)';
ii(ix) = [];
ix = ii;
end
fnix = fn(ix,:);
|
github
|
domingomery/Balu-master
|
Bio_decisionline.m
|
.m
|
Balu-master/InputOutput/Bio_decisionline.m
| 1,513 |
utf_8
|
05d064a41f2cf114260a6d383b1815c9
|
% Bio_decisionline(X,d,op)
%
% Toolbox: Balu
%
% Diaplay a 2D feature space and decision line.
%
% X: Sample data
% d: classification of samples
% op: output of a trained classifier.
%
% Example:
% load datagauss % simulated data (2 classes, 2 features)
% Xn = ['\beta_1';'\beta_2'];
% b(1).name = 'knn'; b(1).options.k = 5; % KNN with 5 neighbors
% b(2).name = 'lda'; b(2).options.p = []; % LDA
% b(3).name = 'svm'; b(3).options.kernel = 4; % rbf-SVM
% op = b;
% op = Bcl_structure(X,d,op);
% close all
% Bio_decisionline(X,d,Xn,op);
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function Bio_decisionline(X,d,Xn,op)
if size(X,2)~=2
error('Bio_decisionline works for two features only.')
end
clf
Bio_plotfeatures(X,d,Xn);
n = length(op);
hold on
ax = axis;
s=0.1;
x = (ax(1)):s:(ax(2));
nx = length(x);
y = (ax(3)):s:(ax(4));
ny = length(y);
rx = ones(ny,1)*x;
ry = y'*ones(1,nx);
XXt = [rx(:) ry(:)];
op = Bcl_structure(X,d,op);
dt = Bcl_structure(XXt,op);
dmin = min(d);
dmax = max(d);
scol = 'ycwkbr';
tcol = 'brgywk';
close all
for k=1:n
figure
Bio_plotfeatures(X,d,Xn);
title(op(k).options.string)
for i=dmin:dmax
ii = find(dt(:,k)==i);
plot(XXt(ii,1),XXt(ii,2),[scol(i-dmin+1) 'o']);
ii = find(d==i);
plot(X(ii,1),X(ii,2),[tcol(i-dmin+1) 'x']);
end
end
|
github
|
domingomery/Balu-master
|
Bio_fmtconv.m
|
.m
|
Balu-master/InputOutput/Bio_fmtconv.m
| 592 |
utf_8
|
56ba969fa5f60391596fb27318ff012b
|
% Bio_fmtconv(fmt1,fmt2)
%
% Toolbox: Balu
% Image format conversion from format fmt1 to fmt2.
%
% This program convert all fmt1 images of current directory to fmt2
% images. fmt1 and fmt2 are strings.
%
% Example:
% Bio_fmtconv('jpg','png') % converts all jpg images into png images
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function Bio_fmtconv(fmt1,fmt2)
f1 = dir(['*.' fmt1]);
t1 = length(fmt1);
n = length(f1);
if n>0
for i=1:n
fi1 = f1(i).name;
I = imread(fi1);
fi2 = [fi1(1:end-t1) fmt2];
imwrite(I,fi2,fmt2);
end
end
|
github
|
domingomery/Balu-master
|
Bio_edgeview.m
|
.m
|
Balu-master/InputOutput/Bio_edgeview.m
| 1,561 |
utf_8
|
97894a03219d2b5a9ad0fb6ad1fc785a
|
% Bio_edgeview(B,E,c,g)
%
% Toolbox: Balu
% Display gray or color image I overimposed by color pixels determined
% by binary image E. Useful to display the edges of an image.
% Variable c is the color vector [r g b] indicating the color to be displayed
% (default: c = [1 0 0], i.e., red)
% Variable g is the number of pixels of the edge lines, default g = 1
%
% Example to display a red edge of a food:
% I = imread('testimg1.jpg'); % Input image
% [R,E] = Bim_segbalu(I); % Segmentation
% Bio_edgeview(I,bwperim(E),[0 1 0],3) % perimeter with 3 pixels will be
% % displayed ingreen ([0 1 0] for [R G B])
%
% D.Mery, PUC-DCC, Apr. 2008-2019
% http://dmery.ing.puc.cl
%
function Bio_edgeview(B,E,cc,g)
if not(exist('cc','var'))
cc = [1 0 0];
end
if not(exist('g','var'))
g = 1;
end
B = double(B);
if max(B(:))>1
B = B/256;
end
if (size(B,3)==1)
[N,M] = size(B);
J = zeros(N,M,3);
J(:,:,1) = B;
J(:,:,2) = B;
J(:,:,3) = B;
B = J;
end
B1 = B(:,:,1);
B2 = B(:,:,2);
B3 = B(:,:,3);
Z = B1==0;
Z = and(Z,B2==0);
Z = and(Z,B3==0);
ii = find(Z==1);
if not(isempty(ii))
B1(ii) = 1/256;
B2(ii) = 1/256;
B3(ii) = 1/256;
end
warning off
E = imdilate(E,ones(g,g));
ii = find(E==1);
B1(ii) = cc(1)*255;
B2(ii) = cc(2)*255;
B3(ii) = cc(3)*255;
Y = double(B);
Y(:,:,1) = B1;
Y(:,:,2) = B2;
Y(:,:,3) = B3;
imshow(uint8(Y*256))
drawnow
warning on
|
github
|
domingomery/Balu-master
|
Bio_plotroc.m
|
.m
|
Balu-master/InputOutput/Bio_plotroc.m
| 910 |
utf_8
|
416a3f90f5d3101cdce92f66261d786e
|
% Bio_plotroc(FPR,TPR,col)
%
% Toolbox: Balu
% Plot ROC curve and fit to an exponetial curve
%
% Example:
%
% th = 3;x = 0:0.05:1; y = 1-exp(-3*x)+randn(1,21)*0.05;
% Bio_plotroc(x,y)
%
% D.Mery, PUC-DCC, Apr. 2013
% http://dmery.ing.puc.cl
%
function [AUC,TPRs,FPRs,TPR05] = Bio_plotroc(x,y,col_line,col_point)
if ~exist('col_line','var')
col_line = 'b';
end
if ~exist('col_point','var')
col_point = 'r.';
end
% clf
plot(x,y,col_point)
ths = fminsearch(@thest,1,[],x,y);
xs = 0:0.005:1;
a = 1/(1-exp(-ths));
ys = a*(1-exp(-ths*xs));
AUC = a*(1-exp(-ths)/ths-1/ths);
hold on
plot(xs,ys,col_line)
d = xs.*xs + (1-ys).*(1-ys);
[~,ii] = min(d);
TPRs = ys(ii(1));
FPRs = xs(ii(1));
ii = find(xs==0.05);
TPR05 = ys(ii(1));
plot(FPRs,TPRs,[col_point(1) '*'])
xlabel('FPR')
ylabel('TPR')
end
function err = thest(th,x,y)
a = 1/(1-exp(-th));
ys = a*(1-exp(-th*x));
err = norm(y-ys);
end
|
github
|
domingomery/Balu-master
|
Bio_segshow.m
|
.m
|
Balu-master/InputOutput/Bio_segshow.m
| 1,424 |
utf_8
|
f50243f7c3b4c583e1508bb678c5d855
|
% Bio_segshow(I,p)
%
% Toolbox: Balu
% Display original image and segmented image
%
% Bimshow display 4 images:
% 1: Original image (matrix I)
% 2: Segmented (using command Bsegbalu)
% 3: High contrast (using command Bsegbalu)
% 4: Edges (using Bedgeview)
%
% p is the parameter used by command Bsegbalu.
%
% Example 1: Segmentation using Balu segmentation algorithm
% I = imread('testimg1.jpg');
% Bio_segshow(I)
% Repeat this examples for images testimg2, testimg3 and testimg4. Last
% test image requires R = Bimshow(I,-0.1) for better results.
%
% Example 2: Segmentation using PCA and with more sensibility
% I = imread('testimg2.jpg');
% Bio_segshow(I,'Bim_segpca',-0.15)
%
% Example 3: The name of the image can be given as argument
% Bio_segshow('testimg4.jpg','Bim_segmaxfisher',-0.1)
%
% See also Bim_segbalu.
%
% D.Mery, PUC-DCC, May 2008
% http://dmery.ing.puc.cl
%
function Bio_segshow(I,segname,p)
if not(exist('segname','var'))
segname = 'Bim_segbalu';
end
if compare(class(I),'char')==0
I = imread(I);
end
if not(exist('p','var'))
p = 0;
end
fsname = ['[R,E,J] = ' segname '(I,p);'];
eval(fsname);
subplot(2,2,1);imshow(I);title('original image')
subplot(2,2,2);imshow(R);title('segmented image')
subplot(2,2,3);imshow(J);title('high contrast image')
subplot(2,2,4);Bio_edgeview(I,imdilate(E,ones(3,3)));title('edge image')
|
github
|
domingomery/Balu-master
|
Bio_drawellipse.m
|
.m
|
Balu-master/InputOutput/Bio_drawellipse.m
| 1,213 |
utf_8
|
78e8e0a8b6677be967be516b6691e0ca
|
% Bio_drawellipse(v,ecol)
%
% Toolbox: Balu
% Draws an ellipse with a(1)x^2 + a(2)xy + a(3)y^2 + a(4)x + a(5)y + a(6) = 0
% ecol is the color of the ellipse
%
% Extracted from http://homepages.inf.ed.ac.uk/rbf/CVonline/
% CVonline: The Evolving, Distributed, Non-Proprietary, On-Line Compendium
% of Computer Vision
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function Bio_drawellipse(v,ecol)
% convert to standard form: ((x-cx)/r1)^2 + ((y-cy)/r2)^2 = 1
% rotated by theta
% v = solveellipse(a);
% draw ellipse with N points
if not(exist('ecol','var'))
ecol = 'b';
end
n = size(v,1);
hold on
for r=1:n
ae = v(r,3);
be = v(r,4);
theta = v(r,5);
mcx = v(r,1);
mcy = v(r,2);
if n>1
text(mcx,mcy,num2str(r))
end
N = 100;
dx = 2*pi/N;
R = [ [ cos(theta) sin(theta)]', [-sin(theta) cos(theta)]'];
X = zeros(N+1);
Y = X;
for i = 1:N
ang = i*dx;
x = ae*cos(ang);
y = be*sin(ang);
d1 = R*[x y]';
X(i) = d1(1) + mcx;
Y(i) = d1(2) + mcy;
end
X(N+1) = X(1);
Y(N+1) = Y(1);
plot(X,Y,ecol)
end
|
github
|
domingomery/Balu-master
|
Bio_maillist.m
|
.m
|
Balu-master/InputOutput/Bio_maillist.m
| 1,572 |
utf_8
|
401aa7f4de4ad67c91080287caf7f9eb
|
% Bio_maillist(mymail,mypassword,mails,subject,heads,body,signature)
%
% Toolbox: Balu
% Send e-mail list
%
% Send an e-mail form mymail to mails, with subject, message, head(s) and
% signature. Ir requires the password of mymail.
%
% mails = {'[email protected]','[email protected]','[email protected]'};
% heads = {'Hi Peter:','Dear Rosa:','Hello Tomas:'};
% message = {'Please do not forget the next meeting on Monday.'};
% subject = {'Next meeting'};
% signature = {'Regards',' ','John',' ','----------------------------------','Prof. John Schmidt','Departmento of Computer Science','University of ...','http://www.usw.edu','----------------------------------'};
% Bio_maillist('[email protected]','johnssszzz12',mails,subject,heads,message,signature);
%
% See also Bio_sendmail.
%
% (c) GRIMA-DCCUC, 2012
% http://grima.ing.puc.cl
function Bio_maillist(mymail,mypassword,mails,subject,heads,body,signature)
n = length(mails);
n1 = length(heads);
if n~=n1
error('number of heads must be equal to number of mails...');
end
nb = length(body);
if and(nb~=1,nb~=n)
error('number of bodies must be equal to 1 or equal to number of mails...');
end
ns = length(subject);
if and(ns~=1,ns~=n)
error('number of subjects must be equal to 1 or equal to number of mails...');
end
if nb==1
msg = body;
end
if ns==1
sbj = char(subject);
end
for i=1:n
if nb>1
msg = body{i};
end
if ns>1
sbj = char(subject{i});
end
Bio_sendmail(mymail,mypassword,mails{i},sbj,[heads{i} ' ' msg ' ' signature]);
end
|
github
|
domingomery/Balu-master
|
Bio_labelregion.m
|
.m
|
Balu-master/InputOutput/Bio_labelregion.m
| 2,719 |
utf_8
|
cd7f29d456b55ec1242ce4df66773c05
|
% [d,D] = Bio_labelregion(I,L,c)
%
% Toolbox: Balu
% User interface to label regions of an image.
%
% I is the original image (color or grayvalue).
% L is a labeled image that indicates the segmented regions of I.
% c is the maximal number of classes.
% d(i) will be the class number of region i.
% D is a binary image with the corresponding labels.
%
% Example:
% I = imread('rice.png');
% I = I(1:70,1:70); % input image
% [L,m] = Bim_segmowgli(I,ones(size(I)),40,1.5); % segmented image
% [d,D] = Bio_labelregion(I,L,3);
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [d,D] = Bio_labelregion(I,L,c)
if size(I,3)==3
J = rgb2gray(I);
else
J = I;
end
close all
cmap = [0 0 1
0 1 0
1 0 0
0 1 1
1 0 1
1 1 0
0 0 0
1 1 1];
cmap = [cmap;cmap;cmap];
figure(2)
Icol = [];
for i=1:c
Icol = [Icol;ones(20,20)*i];
end
imshow(Icol,cmap)
hold on
for i=1:c
text(5,i*20-10,num2str(i))
end
title('Label color')
colorstr = 'bgrcmykwbgrcmykwbgrcmykw';
warning off
figure(1)
subplot(1,2,2)
imshow(I,[])
title('Original image')
hold on
n = max(max(L));
d = zeros(n,1);
i = 1;
D = zeros(size(J));
while(i<=n)
R = zeros(size(J));
kk = find(L==i);
R(kk) = 1;
E = bwperim(R,4);
%subplot(2,2,3)
%Bio_edgeview(J,E);
%title('Class label?')
%figure(3)
subplot(1,2,1)
Bio_edgeview(J,R,[1 1 0]);
title('Class label of yellow region?')
ok = 0;
while(not(ok))
r = input(sprintf('Region %3d/%3d: Class label? (1... %d) or -1 to correct previous: ',i,n,c));
% r = input(['Class label? (1...' num2str(c) ') or -1 to correct previous: ']);
r = round(r);
if not(isempty(r))
if (r==-1)
i = max(i-2,0);
ok = 1;
disp('Correction: enter new choise...')
end
if (r>=1) && (r<=c)
d(i) = r;
D(kk) = r;
ok = 1;
[ii,jj] = find(R==1);
x1 = max(jj);
x2 = min(jj);
y1 = max(ii);
y2 = min(ii);
subplot(1,2,2)
plot([x1 x1 x2 x2 x1],[y1 y2 y2 y1 y1],colorstr(r));
title('Labeled regions')
end
end
if (ok==0)
beep
end
end
i = i + 1;
end
Dc = D+1;
map = [0 0 0;cmap];
subplot(1,2,2)
imshow(Dc,map);
subplot(1,2,1)
Bio_edgeview(J,zeros(size(J)),[1 1 0]);
title('Original image')
|
github
|
domingomery/Balu-master
|
Bio_loadimg.m
|
.m
|
Balu-master/InputOutput/Bio_loadimg.m
| 4,338 |
utf_8
|
a0434706346aa5c8206980dd19f0403b
|
% I = Bloadimg(f,i)
%
% Toolbox: Balu
%
% Load image i of set image defined by structure f
%
% f.path : directory where are the files
% f.extension : extension (eg: 'jpg')
% f.prefix : prefix (eg: 'DSC_')
% f.digits : number of digits (eg:4)
% f.gray : 1 means rgb to gray conversion
% f.subsample : subsampling rate (eg:1 means no subsample)
% f.resize : parameter of imresize, 0 means no imresize
% f.window : image window
% f.negative : negative window
% f.sequence : if seq = [3 2 1], the image for i=1 will be No. 3
% f.imgmin : minimal number of the images (not used by Bloadimg)
% f.imgmax : maximal number of the images (not used by Bloadimg)
% f.show : 1 means display image (deafult = 0)
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.jpg';
% f.prefix = 'testimg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% I = Bloadimg(f,3);
% imshow(I,[])
%
% See Bseq_show.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [I,st] = Bio_loadimg(f,i)
ipath = f.path;
iext = f.extension;
ipre = f.prefix;
if isfield(f,'resize')
re = f.resize;
else
re = 0;
end
if isfield(f,'subsample')
t = f.subsample;
else
t = 1;
end
if isfield(f,'show')
show = f.show;
else
show = 0;
end
if isfield(f,'window')
w = f.window;
else
w = [];
end
if isfield(f,'negative')
neg = f.negative;
else
neg = 0;
end
if isfield(f,'gray')
gr = f.gray;
else
gr = 0;
end
if isfield(f,'sequence')
seq = f.sequence;
else
seq = f.imgmin:f.imgmax;
end
if ~isempty(ipath)
if ipath(end)~='/'
ipath = [ipath '/'];
end
end
%try
% I = f.images(:,:,f.sequence(i)-f.imgmin+1);
%
%catch exception
if isfield(f,'images')
I = f.images(:,:,f.sequence(i)-f.imgmin+1);
% I = f.images(:,:,i-f.imgmin+1);
st = 'from memory';
else
if show
fprintf('Loading image %d...',i);
end
if ipre == '*'
if iext(1)=='.'
iext = iext(2:end);
end
dfiles = [dir([ipath '*.' lower(iext)]); dir([ipath '*.' upper(iext)])];
%tf = cat(1,dfiles.name);
%[tj,ti] = sort(tf);
ti = 1:length(dfiles);
if (i>0)&&(i<=length(dfiles))
st = [ipath dfiles(ti(i)).name];
if ~exist(st,'file')
error('file %s does not exist.',ipath)
end
if compare(upper(iext),'.MAT')==0
st(end-2:end) = 'mat';
s = ['load ' st];
eval(s);
else
I = imread(st);
if show
disp(st)
end
end
else
st = -1;
if i==0
I = length(tf);
else
I = -1;
end
end
else
% if iext(1)~='.'
% iext = ['.' iext]
% end
d = f.digits;
sti = sprintf('0000000%d',seq(i));
sti = sti(end-d+1:end);
st = sprintf('%s%s%s%s',ipath,ipre,sti,iext);
if (exist(st,'file'))
if compare(upper(iext),'.MAT')==0
st(end-2:end) = 'mat';
s = ['load ' st];
eval(s);
else
I = imread(st);
if show
disp(st)
end
end
else
I = -1;
fprintf('%s does not exist.\n',st)
end
end
if length(I(:))>1
if ~isempty(w)
I = I(w(1):w(2),w(3):w(4),:);
end
if size(I,3)==3
if gr
I = rgb2gray(I(1:t:end,1:t:end,:));
else
I = I(1:t:end,1:t:end,:);
end
else
I = I(1:t:end,1:t:end,:);
end
if neg
I = 256.0-double(I);
end
end
I = double(I);
if re>0
I = imresize(I,re);
end
if show
imshow(I/256,[])
end
end
|
github
|
domingomery/Balu-master
|
Bio_latextable.m
|
.m
|
Balu-master/InputOutput/Bio_latextable.m
| 1,315 |
utf_8
|
39dd12c1612493fcdb51ecf1126a51f8
|
%function Bio_latextable(row_names,col_names,fmt,T)
%
% Toolbox: Balu
% Code for a latex table.
%
% row_names is a cell with the names of the rows.
% col_names is a cell with the names of the columns.
% fmt is a cell with the format of each column.
% T is the table.
%
% Example:
%
% col_names = {'cols','col 1','col 2','col 3','col4'};
% row_names = {'row 1','row 2','row 3'};
% fmt = {'%5.2f','%6.4f','%3.1f','%7.4f'};
% T = rand(3,4);
% Bio_latextable(row_names,col_names,fmt,T)
%
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function Bio_latextable(row_names,col_names,fmt,T)
disp('%')
disp('% Latex code starts here:')
disp('%')
[N,M] = size(T);
s = char(ones(M+1,1)*' c ')';
fprintf('\\begin{table}\n')
fprintf('\\caption{Please write caption here.}\n')
fprintf('\\begin{center}\n');
fprintf('\\begin{tabular}{ %s }\n',s(:));
fprintf('\\hline\n');
for j=1:M
fprintf(' %s &',col_names{j});
end
fprintf(' %s \\\\\n',col_names{M+1});
fprintf('\\hline\n');
for i=1:N
fprintf(' %s ',row_names{i});
for j=1:M
s = ['fprintf(' char(39) ' & ' fmt{j} char(39) ',T(i,j));'];
eval(s);
end
fprintf('\\\\\n');
end
fprintf('\\hline\n');
fprintf('\\end{tabular}\n');
fprintf('\\label{Tab:LABEL}\n');
fprintf('\\end{center}\n');
fprintf('\\end{table}\n');
|
github
|
domingomery/Balu-master
|
Bsq_trifocal.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_trifocal.m
| 1,511 |
utf_8
|
88e029db5882015f7975ae21b960c995
|
% F = Bsq_trifocal(P)
%
% Toolbox: Balu
%
% Trifocal tesonsors of a sequence.
%
% P includes the projection matrices of n views as follows:
% Projection Pk = P(k*3-2:k*3,:), for k=1,...,n
%
% T are the fundamental matrices stored as follows:
% T(p,q,r,:) are the trifocal tensors (as 27x1 vector) between views p,
% q and r for p=1:n-2, q=1:p+1:n-1, r=q+1:n
%
% Example:
%
% P1 = rand(3,4); % proyection matrix for view 1
% P2 = rand(3,4); % proyection matrix for view 2
% P3 = rand(3,4); % proyection matrix for view 3
% P4 = rand(3,4); % proyection matrix for view 4
% P5 = rand(3,4); % proyection matrix for view 4
% P6 = rand(3,4); % proyection matrix for view 4
% P = [P1;P2;P3;P4;P5;P6]; % all projection matrices
% T = Bsq_trifocal(P); % fundamental matrices
% T136 = zeros(3,3,3);
% T136(:) = T(1,3,6,:) % Trifocal tensors between views 1-3-6
%
% See also Bmv_fundamental, Bmv_trifocal, Bsq_fundamental.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function T = Bsq_trifocal(P)
mg = size(P,1)/3;
T = zeros(mg,mg,mg,27);
for p=1:mg-2
p0 = 3*p-2;
Pp = P(p0:p0+2,:);
for q=p+1:mg-1
q0 = 3*q-2;
Pq = P(q0:q0+2,:);
for r=q+1:mg
r0 = 3*r-2;
Pr = P(r0:r0+2,:);
Tpqr = Bmv_trifocal(Pp,Pq,Pr);
T(p,q,r,:) = Tpqr(:)';
end
end
end
|
github
|
domingomery/Balu-master
|
Bsq_visualvoc.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_visualvoc.m
| 4,045 |
utf_8
|
3cba9b1709bf6c98fe60dd585555a665
|
% From Spyrou et al 2010:
% ... most of the cmmon visual words, i.e., those with smallest iDF values,
% are not descriminative and their abscence would facilitate the retrieval
% process. On the other hand, the rarest visual words are in most of the
% cases as result of noise and may distract the retrieval process. To
% overcome those problems, a stop list is created that includes the most
% and the least frequent visual words of the image collection.
%
% From Sivic & Zisserman (2003):
% Using a stop list the most frequent visual words that occur in almost all
% images are supressed. The top 5% (of the frequency of visual words over
% all the keyframes) and bottom 10% are stopped.
%
% Term Frequency-Inverse Document Frequency Weighting
% After Sivic & Zisserman (PAMI 2009). See Section 4.1
% Each image of the training database is one 'document'
% The words of each document will be filtered out using a stop list
% (see program asr_stoplist.m).
%
% D.Mery, Notre Dame (2014)
function opvoc = Bsq_visualvoc(Y,options)
top_bound = options.top_bound; % eg. 0.05 top : most frequent word to be stopped
bottom_bound = options.bottom_bound; % eg. 0.10 bottom : less frequent word to be stopped
NV = options.NV; % number of visual words
ixn = options.ixn; % indices of Y rows per document
ii_top = round(NV*(1-top_bound))+1:NV;
ii_bottom = 1:round(NV*bottom_bound);
ii_stop = [ii_bottom ii_top];
show = options.show;
if isfield(options,'newdictionary');
newdictionary = options.newdictionary;
else
newdictionary = 1;
end
if newdictionary == 1
if show>0
fprintf(' computing visual vocabulary from (%dx%d) with %d visual words...',size(Y,1),size(Y,2),NV);
tic
end
% [opvoc.voc,opvoc.ix_voc] = vl_kmeans(Y',NV,'Algorithm','Elkan');
[opvoc.voc,opvoc.ix_voc] = vl_kmeans(Y',NV,'Algorithm','Elkan','NumRepetitions',3);
if show>0
t = toc;
fprintf('in %f sec.\n',t);
end
else
fprintf('Using pre-computed vocabulary with %d visual words...\n',NV);
end
ww = double(opvoc.ix_voc);
N = max(ixn); % # documents in the whole database
nid = zeros(N,NV); % nid(d,i) # occurences of word i in document d
ii = 1:NV; % indices of the words of the vocabulary
for d=1:N
dd = ixn==d; % indices of words in Y of document d
ww_d = ww(dd);
nid(d,:) = hist(ww_d,ii); % # occurrences of word i=1...NV in document d
end
% term frequency nid(d,i) = number of occurences of word i in document d, d=1,...k, t=1,...NV
Ni = sum(nid>0); % document frequency (Ni(i) = # of documents in the collection containing word i, i=1,...NV)
[sNi,jj] = sort(Ni);
opvoc.i_stop = jj(ii_stop);
opvoc.i_go = jj;
opvoc.i_go(ii_stop) = [];
opvoc.i_stop = sort(opvoc.i_stop);
opvoc.i_go = sort(opvoc.i_go);
opvoc.kd_voc = vl_kdtreebuild(opvoc.voc);
opvoc.kd_go = vl_kdtreebuild(opvoc.i_go);
opvoc.voc_stop = opvoc.voc(:,opvoc.i_stop);
opvoc.voc_go = opvoc.voc(:,opvoc.i_go);
opvoc.Ni = Ni;
Ni(opvoc.i_stop) = 0;
opvoc.Ni_go = Ni;
nid_go = nid;
nid_go(:,opvoc.i_stop) = 0;
if options.tfidf > 0
% The following steps are necessary to create a stop list
% the frequencies are computed after eliminating the stop words
nd = sum(nid_go,2); % nd(d): # documents in document d (stop list words are not included)
iDF = log(double(N)./(double(Ni)+1e-20));
opvoc.Vd = Bft_uninorm((nid_go./(nd*ones(1,NV))).*(ones(N,1)*iDF));
end
% similar to Fig. 5 of Sivic & Zisserman (2003)
if show>2
figure(12);clf;
subplot(1,2,1); mesh(nid);
subplot(1,2,2);plot(sNi);ax=axis;
hold on
plot([min(ii_top) min(ii_top)] ,[ax(3) ax(4)],'r:')
plot([max(ii_bottom) max(ii_bottom)],[ax(3) ax(4)],'r:')
end
|
github
|
domingomery/Balu-master
|
Bsq_vocabulary.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_vocabulary.m
| 3,004 |
utf_8
|
4b7cf03feea650d98100b5670be8eb4f
|
% [v,Xcen,H] = Bsq_vocabulary(kp,V,options)
%
% Toolbox: Balu
%
% Visual vocabulary.
%
% kp.des is the descriptions of the keypoint.
% kp.img is a vector containing the number of the image of each keypoint.
% The minimum value of img is allways 1 and the maximum is the number of
% processed images, i.e., f.imgmax-f.imgmin+1.
% V is the number of visual words (clusters).
% options.show display results.
%
% v is the document representation using "term frequency-inverse
% document frequency"
% Xcen are the centroids of the kmeans.
% H are the histograms.
%
%
% Reference:
% Sivic & Zisserman: Efficient visual search for videos cast as text
% retrieval. 31(4):591-606. PAMI 2009.
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.jpg';
% f.prefix = 'testimg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = [256 256];
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% options.show = 1;
% options.descriptor = 'sift';
% options.clean = 0;
% kp = Bsq_des(f,options);
% [v,Xcen,H] = Bsq_vocabulary(kp,100,options);
%
% See also Bsq_vgoogle, Bsq_sort.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [v,Xcen,H,Vnew,kd,N,Ni,Xcen_old,jsel] = Bsq_vocabulary(kp,V,options)
if ~exist('options','var')
options.show = 0;
end
if ~isfield(options,'tfidf')
options.tfidf = 1;
end
if ~isfield(kp,'img')
kp.img = ones(size(kp.des),1);
end
if ~isfield(options,'show')
options.show = 0;
end
show = options.show;
if show
fprintf('Bsq_vocabul: Vector Quantization with %d clusters...\n',V)
end
X = single(kp.des);
Xcen = (vl_kmeans(X',V,'Algorithm','ANN'))';
Xcen_old = Xcen;
if show
disp('Bsq_vocabul: building kd-tree...')
pause(0)
end
kd = vl_kdtreebuild(Xcen');
H = [];
Vnew = V;
v = [];
if options.tfidf == 1
N = max(kp.img);
H = zeros(N,V);
Nd = zeros(N,1);
for d=1:N
h = zeros(1,V);
if show
fprintf('Bsq_vocabul: processing image %d\n',d);
end
ii = find(kp.img==d);
if ~isempty(ii)
Xt = single(kp.des(ii,:));
j = vl_kdtreequery(kd,Xcen',Xt','NumNeighbors',1)';
nd = length(j);
Nd(d) = nd;
for k=1:nd;
h(j(k))=h(j(k))+1;
end
H(d,:) = h;
end
end
H1 = H>0;
hs = mean(H1);
j = hs>0.85;
H(:,j) = [];
jsel = not(j);
Xcen(j,:) = [];
V = size(H,2);
v = zeros(N,V);
Ni = sum(H>0,1);
for d=1:N
nd = Nd(d);
if (nd>0)
for i=1:V
nid = H(d,i);
ti = nid/nd*log(N/Ni(i));
v(d,i) = ti;
end
end
end
Vnew = V;
end
|
github
|
domingomery/Balu-master
|
Bsq_fundamentalSIFT.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_fundamentalSIFT.m
| 1,957 |
utf_8
|
de93354fcb011af6860d6243c67be119
|
% function [Fpq,Bpq] = Bsq_fundamentalSIFT(kp,p,q)
%
% Toolbox: Balu
%
% Fundamental matrix between two views p and q of a sequence using SIFT
% descriptors kp.
%
% Example: case when keypoints must be extracted
%
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.png';
% f.prefix = 'X';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% options.show = 1;
% options.descriptor = 'harris+sift';
% options.clean = 0;
% kp = Bsq_des(f,options);
% F13 = Bsq_fundamentalSIFT(kp,1,3);
% F14 = Bsq_fundamentalSIFT(kp,1,4);
% figure(1); Bio_imgshow(f,1,[]); hold on
% figure(2); Bio_imgshow(f,3,[]); hold on
% figure(3); Bio_imgshow(f,4,[]); hold on
% while(1)
% figure(1);
% disp('click a point in Figure 1...') % click
% p = vl_click; m1 = [p(2) p(1) 1]';
% plot(p(1),p(2),'g+')
% figure(2)
% Bmv_epiplot(F13,m1)
% figure(3)
% Bmv_epiplot(F14,m1)
% end
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [Fpq,Bpq] = Bsq_fundamentalSIFT(kp,p,q)
img = kp.img;
des = kp.des;
fra = kp.fra;
% Fundamental matrix for image p and q using SIFT matching and RANSAC
ip = find(img==p);
iq = find(img==q);
dp = des(ip,:);
dq = des(iq,:);
frap = fra(ip,:);
fraq = fra(iq,:);
[matches, scores] = vl_ubcmatch(dp', dq');
ii = find(scores<17000);
xp = frap(matches(1,ii),[2 1]);
xq = fraq(matches(2,ii),[2 1]);
[Fpq, i] = Bmv_fundamentalRANSAC(xp',xq');
inl = ii(i);
matchp = matches(1,inl);
matchq = matches(2,inl);
Bpq = [ip(matchp) iq(matchq)];
|
github
|
domingomery/Balu-master
|
Bsq_movie.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_movie.m
| 1,868 |
utf_8
|
40172c3d3a79e3c6b61ff0ff1899b605
|
% M = Bsq_show(f,map,k)
%
% Toolbox: Balu
%
% Display a movie of an image sequence defined by structure f.
%
% map is the map of the image, if not given will be used "[]".
%
% f.path : directory where are the files
% f.extension : extension (eg: 'jpg')
% f.prefix : prefix (eg: 'DSC_')
% f.digits : number of digits (eg:4)
% f.gray : 1 means rgb to gray conversion
% f.subsample : subsampling rate (eg:1 means no subsample)
% f.resize : parameter of imresize, 0 means no imresize
% f.window : image window
% f.negative : negative window
% f.sequence : if seq = [3 2 1], the image for i=1 will be No. 3
% f.imgmin : minimal number of the images (not used by Bio_loadimg)
% f.imgmax : maximal number of the images (not used by Bio_loadimg)
%
% Iseq is the output image with all images of the sequence.
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.jpg';
% f.prefix = 'testimg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = [100 100];
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% M = Bsq_movie(f);
%
% See Bio_loadimg, Bsq_show, movie, getframe.
%
% (c) D.Mery, PUC-DCC, 2012
% http://dmery.ing.puc.cl
function M = Bsq_movie(f,map,k)
clf
% M = [];
j = 0;
for i=f.imgmin:f.imgmax
j = j+1;
% s = f.sequence(i)
s = i;
II = Bio_loadimg(f,s);
if exist('map','var')
if exist('k','var')
if k>0
imshow(k*II,map)
end
else
imshow(II,map)
end
else
imshow(II,[])
end
M(j) = getframe;
enterpause
end
% movie(M)
|
github
|
domingomery/Balu-master
|
Bsq_des.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_des.m
| 9,491 |
utf_8
|
aab8368709486b1a39dd2a711d00ad9c
|
% kp = Bsq_des(f,options)
%
% Toolbox: Balu
%
% Description of a sequence.
%
% f is the structure that defines the sequence
%
% f.path : directory where are the files
% f.extension : extension (eg: 'jpg')
% f.prefix : prefix (eg: 'DSC_')
% f.digits : number of digits (eg:4)
% f.gray : 1 means rgb to gray conversion
% f.subsample : subsampling rate (eg:1 means no subsample)
% f.resize : parameter of imresize, 0 means no imresize
% f.window : image window
% f.negative : negative window
% f.sequence : if seq = [3 2 1], the image for i=1 will be No. 3
% f.imgmin : minimal number of the images (not used by Bio_loadimg)
% f.imgmax : maximal number of the images (not used by Bio_loadimg)
%
% options.descriptor can be: 'sift', 'surf', 'sift-plus', 'phow',
% 'harris', 'harris-defects', 'harris+sift', 'mser', 'clp',
% 'blops', 'clp+harris'.
%
% options.show display results
% options.param are paremeters of the method (if any).
% options.clean is the name of the function used to eliminate keypoints
% outside of the object of interest, this function is a tailored
% function and returns a binary image with '1' object and '0' background
% options.clean = 0 means no cleaning
%
% kp (keypoints) is a structure with the following fields:
% kp.fra and kp.des are the frames and descriptions of each keypoint.
% kp.img is a vector containing the number of the image of each keypoint.
% The minimum value of img is allways 1 and the maximum is the number of
% processed images, i.e., f.imgmax-f.imgmin+1
% kp.ilu is a look up table of the real number of the images (see example).
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.prefix = 'testimg';
% f.extension = '.jpg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = [256 256];
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 5;
% f.imgmax = 6;
% options.show = 1;
% options.descriptor = 'sift';
% options.clean = 0;
%
% kp = Bsq_des(f,options);
%
% % The frames and description of image 5 are:
%
% i = find(kp.ilu==5);
% j = find(kp.img==i);
% f5 = kp.fra(j,:);
% d5 = kp.des(j,:);
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function kp = Bsq_des(f,options)
imin = f.imgmin; % first image
imax = f.imgmax; % last image
i0 = 0;
img = [];
fra = [];
des = [];
ilu = [];
show = options.show;
method = lower(options.descriptor);
for i=imin:imax
if show
fprintf('Bsq_des : processing image %5d...',i)
end
J = Bio_loadimg(f,i);
if length(J(:))>1
i0 = i0+1;
im = single(J);
switch method
case 'sift'
[frames, descrs] = vl_sift(im) ;
case 'surf'
Options.verbose = false;
Options.tresh=0.5;
Options.octaves = 5;
Options.init_sample = 2;
Ipts=OpenSurf(J,Options);
x = cat(1,Ipts.x);
y = cat(1,Ipts.y);
s = cat(1,Ipts.scale);
o = cat(1,Ipts.orientation);
frames = [x';y';s';o'];
n = length(Ipts);
descrs = zeros(64,n);
descrs(:) = cat(1,Ipts.descriptor);
% l = cat(1,Ipts.laplacian);
% ii = l==0;
ii = frames(3,:)<4;
frames = frames(:,ii);
descrs = descrs(:,ii);
case 'sift-plus'
H = vl_harris( vl_imsmooth(J,3),7);
idx = vl_localmax(H,0.3) ;
[y,x] = ind2sub(size(im),idx);
n = length(x);
frames = [x; y; 20*ones(1,n);zeros(1,n)];
% [fhar2, dhar2] = vl_sift(im,'frames',frames) ;
% [fsift, dsift] = vl_sift(im,'EdgeThresh',50) ;
% frames = [fhar2 fsift];
% descrs = [dhar2 dsift];
[fhar2, dhar2] = vl_sift(im,'frames',frames) ;
%[fsift, dsift] = vl_sift(im,'EdgeThresh',50) ;
frames = [fhar2];
descrs = [dhar2];
case 'sifts'
[frames, descrs] = BsiftmatchS(J,options.param.Iref,options.param.t1,options.param.t2,0) ;
case 'phow'
[frames, descrs] = vl_phow(im);
case 'harris'
H = vl_harris( vl_imsmooth(J,3),3);
idx = vl_localmax(H ) ;
[y,x] = ind2sub( size(im), idx );
n = length(x);
frames = [x; y; 20*ones(1,n) ; zeros(1,n)];
[frames, descrs] = vl_sift(im,'frames',frames) ;
case 'harris-defects'
% p = [3 0.2 5 0.5 3];
p = [13 2 5 2 13];
J1 = vl_imsmooth(J,p(5));
im = single(J1);
H = vl_harris(J1,p(1));
idx = vl_localmax(H,p(2));
[y1,x1] = ind2sub(size(im),idx);
H = vl_harris(J,p(3));
idx = vl_localmax(H,p(4));
[y2,x2] = ind2sub(size(im),idx);
x = [x1 x2];
y = [y1 y2];
n = length(x);
frames = [x; y; 5*ones(1,n) ; zeros(1,n)];
[frames, descrs] = vl_sift(im,'frames',frames) ;
case {'harris+sift','sift+harris'}
H = vl_harris( vl_imsmooth(J,3),3);
idx = vl_localmax(H) ;
[y,x] = ind2sub(size(im),idx);
n = length(x);
frames = [x; y; 20*ones(1,n);zeros(1,n)];
[fhar2, dhar2] = vl_sift(im,'frames',frames);
[fsift, dsift] = vl_sift(im) ;
frames = [fhar2 fsift];
descrs = [dhar2 dsift];
case 'mser'
[r,frames] = vl_mser(uint8(J),'MinDiversity',0.7,'MaxVariation',0.2,'Delta',10) ;
% [r,frames] = vl_mser(uint8(J),'MinDiversity',options.param(1),'MaxVariation',options.param(2),'Delta',options.param(3)) ;
frames = vl_ertr(frames) ;
[frames, descrs] = vl_sift(im,'frames',frames(1:4,:));
case 'segmser'
frames = Bim_segmser(J,options.param);
[frames, descrs] = vl_sift(im,'frames',frames);
case 'razor'
frames = RazorDetector(J,options.param);
if ~isempty(frames)
[frames, descrs] = vl_sift(im,'frames',frames);
end
case 'gun'
frames = GunDetector(J,options.param);
if ~isempty(frames)
[frames, descrs] = vl_sift(im,'frames',frames);
end
case 'clp'
[Y,feat] = Bim_segdefects(J);
n = size(feat,1);
fclp = [feat(:,[2 1 3]) zeros(n,1)]';
[fclp, dclp] = vl_sift(im,'frames',fclp);
frames = fclp;
descrs = dclp;
case 'blops'
% options.param = [13 20 0 50 15 200 0.5 1.2]; % para la gillette
% options.param = [17 15 0 100 15 2000 0 0]; % para la manzana
[Y,feat] = Bim_segblops(J,options.param);
n = size(feat,1);
fblops = [feat(:,[2 1 3]) zeros(n,1)]';
[fblops, dblops] = vl_sift(im,'frames',fblops);
frames = fblops;
descrs = dblops;
case 'clp+harris'
H = vl_harris( vl_imsmooth(J,3),3);
idx = vl_localmax(H) ;
[y,x] = ind2sub(size(im),idx);
n = length(x);
frames = [x; y; 10*ones(1,n);zeros(1,n)];
[fhar1, dhar1] = vl_sift(im,'frames',frames) ;
[Y,feat] = segdefects(J);
n = size(feat,1);
fclp = [feat(:,[2 1 3]) zeros(n,1)]';
[fclp, dclp] = vl_sift(im,'frames',fclp) ;
[fsift, dsift] = vl_sift(im) ;
frames = [fclp fhar1 fsift];
descrs = [dclp dhar1 dsift];
end
if ~isfield(options,'clean')
options.clean = 0;
end
if sum(options.clean) > 0
R = feval(options.clean,J);
s = size(J);
x = round(frames(1,:))';
y = round(frames(2,:))';
ix = sub2ind(s,y,x);
t = R(ix);
j = t==1;
frames = frames(:,j);
descrs = descrs(:,j);
end
if show
clf
imshow(J,[]);
pause(0);
x = frames(1,:);
y = frames(2,:);
hold on;
plot(x,y,'.')
fprintf('... %6d keypoints\n',length(x(:)))
drawnow
end
if ~isempty(frames)
fra = [fra; frames'];
des = [des; descrs'];
img = [img; i0*ones(size(frames,2),1)];
ilu = [ilu; i];
end
else
error('Bsq_des : image not found.');
end
end
kp.fra = fra;
kp.des = des;
kp.img = img;
kp.ilu = ilu;
|
github
|
domingomery/Balu-master
|
Bsq_sort.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_sort.m
| 4,689 |
utf_8
|
b7e6b7557c3924943e4e2fa32c851ce0
|
% s = Bsq_sort(kp,f,options)
% [kp_new,files_new] = Bsq_sort(kp,files,options)
%
% Toolbox: Balu
%
% Sort an image sequence.
%
% f is the structure that defines the sequence
%
% f.path : directory where are the files
% f.extension : extension (eg: 'jpg')
% f.prefix : prefix (eg: 'DSC_')
% f.digits : number of digits (eg:4)
% f.gray : 1 means rgb to gray conversion
% f.subsample : subsampling rate (eg:1 means no subsample)
% f.resize : parameter of imresize, 0 means no imresize
% f.window : image window
% f.negative : negative window
% f.sequence : if seq = [3 2 1], the image for i=1 will be No. 3
% f.imgmin : minimal number of the images (not used by Bloadimg)
% f.imgmax : maximal number of the images (not used by Bloadimg)
%
% kp are the keypoints of the sequence (extracted with Bsq_des or any
% similar procedure). If kp is empty the kypoints will be extracted with
% Bsq_des using 'harris+sift' method (see help Bsq_des).
%
% options.show display results
% options.indexonly 1: returns only s
% options.indexonly 0: returns kp_new and files_new, they are the
% corresponding keypoints and files sequence for the sorted images
%
% s is the sorted sequence.
%
% Example 1: case when keypoints must be extracted
%
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.png';
% f.prefix = 'X';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% options.show = 1;
% options.indexonly = 0;
% options.descriptor = 'harris+sift';
% options.clean = 0;
% [kp,f2] = Bsq_sort([],f,options);
% Bsq_show(f); title('unsorted sequence')
% figure
% Bsq_show(f2); title('sorted sequence')
%
%
% Example 2: case when keypoints are already extracted
%
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.png';
% f.prefix = 'X';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% options.show = 1;
% options.indexonly = 0;
% options.descriptor = 'harris+sift';
% kp = Bsq_des(f,options);
% figure
% Bsq_show(f); title('unsorted sequence')
% [kp2,f2] = Bsq_sort(kp,f,options);
% figure
% Bsq_show(f2); title('sorted sequence')
%
% Example 3: only the new indices are riquired
%
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.png';
% f.prefix = 'X';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% options.show = 0;
% options.indexonly = 1;
% options.descriptor = 'harris+sift';
% kp = Bsq_des(f,options);
% s = Bsq_sort([],f,options)
%
% See also Bsq_vgoogle, Bsq_des.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [kp_new,files_new] = Bsq_sort(kp,files,options)
f = files;
show = options.show;
if isempty(kp)
kp = Bsq_des(f,options);
end
v = Bsq_vocabulary(kp,200,options);
n = max(kp.img);
H = zeros(n,n);
for i=1:n
q = kp.ilu==(i+f.imgmin-1);
vq = v(q,:)';
[rk,j] = Bfa_vecsimilarity(vq,v);
H(i,j) = rk;
end
spmax = 0;
for j=1:n
i0 = j;
sj = zeros(n,1);
sj(1) = i0;
w = ones(n,1);
w(i0) = 0;
sp = 0;
for i=2:n
h = w.*H(:,i0);
[p,q] = max(h);
i0 = q;
sj(i) = q;
w(q) = 0;
sp = sp+p;
end
if sp>spmax
spmax = sp;
s = sj;
end
end
if show
figure
Bsq_show(f);title('unsorted sequence')
f.sequence(f.imgmin:f.imgmax) = s+f.imgmin-1;
figure
Bsq_show(f);title('sorted sequence')
end
if options.indexonly
kp_new = s;
files_new = (sum(H)-1)/(n-1); % confidence, small values are not similar enough
else
kp_new = kp;
n = max(kp.img);
for i=1:n
ii = kp.img==s(i);
kp_new.img(ii) = i;
end
files_new = f;
files_new.sequence(files.imgmin:files.imgmax) = s+files.imgmin-1;
end
|
github
|
domingomery/Balu-master
|
Bsq_load.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_load.m
| 1,030 |
utf_8
|
d3512600879f06cd4d8637b6ccfbeb72
|
% f_new = Bsq_load(f)
%
% Toolbox: Balu
%
% Load an image sequence.
%
% f is a file structure (see Bio_loadimg for details)
%
% f_new includes a fild called f_new.images with an array NxMxm for a
% sequence with NxM images
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.jpg';
% f.prefix = 'testimg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% f = Bsq_load(f);
% Ij = f.images;
% imshow(Ij(:,:,3),[])
%
% See also Bload_img.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function f_new = Bsq_load(f)
nimg = f.imgmax-f.imgmin+1;
I = Bio_loadimg(f,f.imgmin);
[N,M] = size(I);
Ij = zeros(N,M,nimg);
Ij(:,:,1) = I;
for j=f.imgmin+1:f.imgmax
Ij(:,:,j-f.imgmin+1) = Bio_loadimg(f,j);
end
f_new = f;
f_new.images = Ij;
|
github
|
domingomery/Balu-master
|
Bsq_vgoogle.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_vgoogle.m
| 2,607 |
utf_8
|
0362e101680bbe07d55060aa36e95924
|
% [rk,j] = Bsq_vgoogle(f,i,ilu,v,show)
%
% Toolbox: Balu
%
% Search sequence images similar to image i.
%
% f is the structure that defines the sequence
%
% f.path : directory where are the files
% f.extension : extension (eg: 'jpg')
% f.prefix : prefix (eg: 'DSC_')
% f.digits : number of digits (eg:4)
% f.gray : 1 means rgb to gray conversion
% f.subsample : subsampling rate (eg:1 means no subsample)
% f.resize : parameter of imresize, 0 means no imresize
% f.window : image window
% f.negative : negative window
% f.sequence : if seq = [3 2 1], the image for i=1 will be No. 3
% f.imgmin : minimal number of the images (not used by Bio_loadimg)
% f.imgmax : maximal number of the images (not used by Bio_loadimg)
%
% ilu is a look up table of the real number of the images (see example),
% if ilu is empty ilu will be f.imgmin:f.imgmax.
% v is the document representation using "term frequency-inverse
% document frequency"
% show display results
%
% Reference:
% Sivic & Zisserman: Efficient visual search for videos cast as text
% retrieval. 31(4):591-606. PAMI 2009.
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.jpg';
% f.prefix = 'testimg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = [256 256];
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% options.show = 1;
% options.descriptor = 'sift';
% options.clean = 0;
% kp = Bsq_des(f,options);
% v = Bsq_vocabulary(kp,100,options);
% [rk,j] = Bsq_vgoogle(f,5,[],v,1); % similar images to image 5
%
% See also Bsq_sort, Bsq_vocabulary.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [rk,j] = Bsq_vgoogle(f,i,ilu,v,show)
if isempty(ilu)
ilu = f.imgmin:f.imgmax;
end
vq = v(ilu==i,:)';
[rk,j] = Bfa_vecsimilarity(vq,v);
if ~exist('show','var')
show = 0;
end
if show
disp('Finding similar images:')
close all
figure(1)
imshow(Bio_loadimg(f,i),[])
title(sprintf('query image %d',i))
m = 4;
for k=m+1:-1:2
figure(k)
imshow(Bio_loadimg(f,ilu(j(k))),[]);
s = sprintf('Figure %d: image %d (Similarity with image %d: %f)',k,ilu(j(k)),i,rk(k));
title(s)
disp(s)
end
figure(1)
disp('Figure 1: Query image')
end
|
github
|
domingomery/Balu-master
|
Bsq_fundamental.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_fundamental.m
| 1,824 |
utf_8
|
539819854dbc3fda7b210c7507b9989d
|
% F = Bsq_fundamental(P)
%
% Toolbox: Balu
%
% Fundamental matrices of a sequence.
%
% P includes the projection matrices of n views as follows:
% Projection Pk = P(k*3-2:k*3,:), for k=1,...,n
%
% F are the fundamental matrices stored as follows:
% F(p,q,:) is the Fundamental matrix (as 9x1 vector) between view p and
% view q, for p=1:n-1 and q=1:p+1:n
%
% Example:
%
% M = [1 2 3 1]'; % 3D point (X=1,Y=2,Z=3)
% P1 = rand(3,4); % proyection matrix for view 1
% P2 = rand(3,4); % proyection matrix for view 2
% P3 = rand(3,4); % proyection matrix for view 3
% P4 = rand(3,4); % proyection matrix for view 4
% m1 = P1*M; m1=m1/m1(3); % proyection point in view 1
% m2 = P2*M; m2=m2/m2(3); % proyection point in view 2
% m3 = P3*M; m3=m3/m3(3); % proyection point in view 3
% m4 = P4*M; m4=m4/m4(3); % proyection point in view 4
% P = [P1;P2;P3;P4]; % all projection matrices
% F = Bsq_fundamental(P); % fundamental matrices
% F13 = zeros(3,3);
% F13(:) = F(1,3,:); % Fundamental matrix between views 1-3
% m3'*F13*m1 % epipolar constraint must be zero
% F24 = zeros(3,3);
% F24(:) = F(2,4,:); % Fundamental matrix between views 2-4
% m4'*F24*m2 % epipolar constraint must be zero
%
% See also Bmv_fundamental, Bmv_trifocal, Bsq_trifocal.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function F = Bsq_fundamental(P)
n = size(P,1)/3;
F = zeros(n,n,9);
for p=1:n-1
i0 = 3*p-2;
Pi = P(i0:i0+2,:);
for q=p+1:n
j0 = 3*q-2;
Pj = P(j0:j0+2,:);
Fij = Bmv_fundamental(Pi,Pj);
F(p,q,:) = Fij(:)';
end
end
|
github
|
domingomery/Balu-master
|
Bsq_multifundamental.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_multifundamental.m
| 2,250 |
utf_8
|
1d51291e3817f5d620f4f0fc99afcf88
|
% function function [F,Bo] = Bsq_multifundamental(kp,options)
%
% Toolbox: Balu
%
% (No calibrated) Multiple fundamental matrices of a sequence
%
% Example: case when keypoints must be extracted
%
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.png';
% f.prefix = 'X';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = 0;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
%
% op1.show = 1;
% op1.descriptor = 'harris+sift';
% op1.clean = 0;
% op2.img1 = 1;
% op2.img2 = 4;
% op2.m = 3;
% op2.show = 1;
%
% kp = Bsq_des(f,op1);
% Fo = Bsq_fmulti(kp,op2);
% ii = and(Fo(:,1)==1,Fo(:,2)==3);
% F13 = zeros(3,3); F13(:) = Fo(ii,3:11);
% ii = and(Fo(:,1)==1,Fo(:,2)==4);
% F14 = zeros(3,3); F14(:) = Fo(ii,3:11);
%
% close all
% figure(1); Bio_imgshow(f,1,[]); hold on
% figure(2); Bio_imgshow(f,3,[]); hold on
% figure(3); Bio_imgshow(f,4,[]); hold on
% while(1)
% figure(1);
% disp('click a point in Figure 1...') % click
% p = vl_click; m1 = [p(2) p(1) 1]';
% plot(p(1),p(2),'g+')
% figure(2)
% Bmv_epiplot(F13,m1);
% figure(3)
% Bmv_epiplot(F14,m1);
% end
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [F,Bo] = Bsq_multifundamental(kp,options)
img1 = options.img1;
img2 = options.img2;
show = options.show;
m = options.m;
N = max(kp.img);
F = zeros(10000,11);
j = 0;
Bo = zeros(20000,2);
i = 1;
if show
disp('Computing Fundamental Matrices...')
end
for p = img1:img2
for q = p+1:min([p+m N])
j = j+1;
if show
fprintf('F(%d,%d)...',p,q)
end
[Fpq,Bpq] = Bsq_fundamentalSIFT(kp,p,q);
F(j,:) = [p q Fpq(:)'];
in = size(Bpq,1);
Bo(i:i+in-1,:) = Bpq;
i = i+in;
end
end
F = F(1:j,:);
Bo = Bo(1:i-1,:);
if show
fprintf('\n');
end
|
github
|
domingomery/Balu-master
|
Bsq_visualvoc_bak.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_visualvoc_bak.m
| 4,007 |
utf_8
|
ad6bb77f38057b5193c205fc35b269ab
|
% From Spyrou et al 2010:
% ... most of the cmmon visual words, i.e., those with smallest iDF values,
% are not descriminative and their abscence would facilitate the retrieval
% process. On the other hand, the rarest visual words are in most of the
% cases as result of noise and may distract the retrieval process. To
% overcome those problems, a stop list is created that includes the most
% and the least frequent visual words of the image collection.
%
% From Sivic & Zisserman (2003):
% Using a stop list the most frequent visual words that occur in almost all
% images are supressed. The top 5% (of the frequency of visual words over
% all the keyframes) and bottom 10% are stopped.
%
% Term Frequency-Inverse Document Frequency Weighting
% After Sivic & Zisserman (PAMI 2009). See Section 4.1
% Each image of the training database is one 'document'
% The words of each document will be filtered out using a stop list
% (see program asr_stoplist.m).
%
% D.Mery, Notre Dame (2014)
function options = Bsq_visualvoc(Y,options)
top_bound = options.top_bound; % eg. 0.05 top : most frequent word to be stopped
bottom_bound = options.bottom_bound; % eg. 0.10 bottom : less frequent word to be stopped
NV = options.NV; % number of visual words
ixn = options.ixn; % indices of Y rows per document
ii_top = round(NV*(1-top_bound))+1:NV;
ii_bottom = 1:round(NV*bottom_bound);
ii_stop = [ii_bottom ii_top];
show = options.show;
if isfield(options,'newdictionary');
newdictionary = options.newdictionary;
else
newdictionary = 1;
end
if newdictionary == 1
if show>0
fprintf(' computing visual vocabulary from (%dx%d) with %d visual words...',size(Y,1),size(Y,2),NV);
tic
end
[options.voc,options.ix_voc] = vl_kmeans(Y',NV,'Algorithm','Elkan');
if show>0
t = toc;
fprintf('in %f sec.\n',t);
end
else
fprintf('Using pre-computed vocabulary with %d visual words...\n',NV);
end
ww = double(options.ix_voc);
N = max(ixn); % # documents in the whole database
nid = zeros(N,NV); % nid(d,i) # occurences of word i in document d
ii = 1:NV; % indices of the words of the vocabulary
for d=1:N
dd = ixn==d; % indices of words in Y of document d
ww_d = ww(dd);
nid(d,:) = hist(ww_d,ii); % # occurrences of word i=1...NV in document d
end
% term frequency nid(d,i) = number of occurences of word i in document d, d=1,...k, t=1,...NV
Ni = sum(nid>0); % document frequency (Ni(i) = # of documents in the collection containing word i, i=1,...NV)
[sNi,jj] = sort(Ni);
options.i_stop = jj(ii_stop);
options.i_go = jj;
options.i_go(ii_stop) = [];
options.i_stop = sort(options.i_stop);
options.i_go = sort(options.i_go);
options.kd_voc = vl_kdtreebuild(options.voc);
options.kd_go = vl_kdtreebuild(options.i_go);
options.voc_stop = options.voc(:,options.i_stop);
options.voc_go = options.voc(:,options.i_go);
options.Ni = Ni;
Ni(options.i_stop) = 0;
options.Ni_go = Ni;
nid_go = nid;
nid_go(:,options.i_stop) = 0;
if options.tfidf > 0
% The following steps are necessary to create a stop list
% the frequencies are computed after eliminating the stop words
nd = sum(nid_go,2); % nd(d): # documents in document d (stop list words are not included)
iDF = log(double(N)./(double(Ni)+1e-20));
options.Vd = Bft_uninorm((nid_go./(nd*ones(1,NV))).*(ones(N,1)*iDF));
end
% similar to Fig. 5 of Sivic & Zisserman (2003)
if show>2
figure(12);clf;
subplot(1,2,1); mesh(nid);
subplot(1,2,2);plot(sNi);ax=axis;
hold on
plot([min(ii_top) min(ii_top)] ,[ax(3) ax(4)],'r:')
plot([max(ii_bottom) max(ii_bottom)],[ax(3) ax(4)],'r:')
end
|
github
|
domingomery/Balu-master
|
Bsq_show.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_show.m
| 2,105 |
utf_8
|
a56ba86964f8832cd23477ff7ef1ece0
|
% Iseq = Bsq_show(f,n,map,k)
%
% Toolbox: Balu
%
% Display an image sequence defined by structure f.
%
% n is the number of images per row (default is the number of images of
% the sequence).
%
% map is the map of the image, if not given will be used "[]".
%
% f.path : directory where are the files
% f.extension : extension (eg: 'jpg')
% f.prefix : prefix (eg: 'DSC_')
% f.digits : number of digits (eg:4)
% f.gray : 1 means rgb to gray conversion
% f.subsample : subsampling rate (eg:1 means no subsample)
% f.resize : parameter of imresize, 0 means no imresize
% f.window : image window
% f.negative : negative window
% f.sequence : if seq = [3 2 1], the image for i=1 will be No. 3
% f.imgmin : minimal number of the images (not used by Bio_loadimg)
% f.imgmax : maximal number of the images (not used by Bio_loadimg)
%
% Iseq is the output image with all images of the sequence.
%
% Example:
% f.path = ''; % Balu directory as path or current directory
% f.extension = '.jpg';
% f.prefix = 'testimg';
% f.digits = 1;
% f.gray = 1;
% f.subsample = 1;
% f.resize = [100 100];
% f.window = [];
% f.negative = 0;
% f.sequence = 1:6;
% f.imgmin = 1;
% f.imgmax = 6;
% Bsq_show(f,3);
%
% See Bio_loadimg.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function Iseq = Bsq_show(f,n,map,k)
if (~exist('n','var'))
n = f.imgmax-f.imgmin+1;
end
t = 0;
II = [];
Iseq = [];
for i=f.imgmin:f.imgmax
s = f.sequence(i);
s = i;
Ii = Bio_loadimg(f,s);
Iseq = [Iseq Ii];
t = t + 1;
if (t==n)
II = [II; Iseq];
t = 0;
Iseq = [];
end
end
if t>0
Iseq = [Iseq zeros(size(Ii,1),size(Ii,2)*(n-t))];
II = [II; Iseq];
end
Iseq = II;
if exist('map','var')
if exist('k','var')
if k>0
imshow(k*II,map)
end
else
imshow(II,map)
end
else
imshow(II,[])
end
|
github
|
domingomery/Balu-master
|
Bsq_stoplist.m
|
.m
|
Balu-master/SequenceProcessing/Bsq_stoplist.m
| 2,649 |
utf_8
|
9e2d1e8f1699640d23b2f43116850843
|
% From Spyrou et al 2010:
% ... most of the common visual words, i.e., those with smallest iDF values,
% are not discriminative and their absence would facilitate the retrieval
% process. On the other hand, the rarest visual words are in most of the
% cases as result of noise and may distract the retrieval process. To
% overcome those problems, a stop list is created that includes the most
% and the least frequent visual words of the image collection.
%
% From Sivic & Zisserman (2003):
% Using a stop list the most frequent visual words that occur in almost all
% images are suppressed. The top 5% (of the frequency of visual words over
% all the keyframes) and bottom 10% are stopped.
function [Voc,ix,i_stop,i_go,kd,kd_go] = Bsq_stoplist(Y,options)
top_bound = options.top_bound; % eg. 0.05 top : most frequent word to be stopped
bottom_bound = options.bottom_bound; % eg. 0.10 bottom : less frequent word to be stopped
NV = options.NV; % number of visual words
ixn = options.ixn; % indices of Y rows per document
ii_top = round(NV*(1-top_bound))+1:NV;
ii_bottom = 1:round(NV*bottom_bound);
ii_stop = [ii_bottom ii_top];
show = options.show;
if isfield(options,'newdictionary');
newdictionary = options.newdictionary;
else
newdictionary = 1;
end
if newdictionary == 1
fprintf('Computing vocabulary of Y (%dx%d) with %d visual words...',size(Y,1),size(Y,2),NV);
tic
[Voc,ix] = vl_kmeans(Y',NV,'Algorithm','Elkan');
t = toc;
fprintf('in %f sec.\n',t);
else
fprintf('Using pre-computed vocabulary with %d visual words...\n',NV);
Voc = options.voc;
ix = options.ix_voc;
end
ww = double(ix);
N = max(ixn);
TF = zeros(N,NV);
for i=1:N
ii = ixn==i;
ww_i = ww(ii);
TF(i,:) = hist(ww_i,1:NV);
end
% term frequency TF(d,t) = number of occurrences of word t in document d, d=1,...k, t=1,...NV
DF = sum(TF>0); % document frequency (DF(t) = number of documents in the collection that contain a word t, t=1,...NV)
% Not necessary to create a stop list
% iDF = log(N./(DF));
% TFIDF = TF.*(ones(k,1)*iDF);
% [ii,jj] = sort(iDF,'descend');
[~,jj] = sort(DF);
i_stop = jj(ii_stop);
i_go = jj; i_go(ii_stop) = [];
kd = vl_kdtreebuild(Voc);
kd_go = vl_kdtreebuild(i_go);
% V_stop = Voc(:,j_stop);
% V_go = Voc(:,j_go);
% like Fig. 5 of Sivic & Zisserman (2003)
if show
figure(12);clf;
subplot(1,3,1); mesh(TF);
subplot(1,3,2);plot(sort(DF,'descend'));ax=axis;
subplot(1,3,3);plot(sort(DF(i_go),'descend'));axis(ax)
end
|
github
|
domingomery/Balu-master
|
Bfx_moments.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_moments.m
| 1,948 |
utf_8
|
653e002ee885a61555adb5dd36829228
|
% [X,Xn] = Bfx_moments(R,options)
%
% Toolbox: Balu
%
% Extract moments and central moments.
%
% options.show = 1 display mesagges.
% options.central = 1 for central moments, and 0 for normal moments
% options.rs = n x 2 matrix that contains the indices of the
% moments to be extracted (see example).
%
% X is vector that contains the n moments.
% Xn is the list of the n feature names.
%
% Example (Centroid of a region)
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% imshow(R);
% options.show = 1;
% options.central = 0;
% options.rs = [0 0; 1 0; 0 1];
% [X,Xn] = Bfx_moments(R,options);
% ic = X(2)/X(1);
% jc = X(3)/X(1);
% hold on
% plot(jc,ic,'rx')
%
% See also Bfx_basicgeo, Bfx_gupta, Bfx_fitellipse, Bfx_flusser,
% Bfx_hugeo.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function [X,Xn] = Bfx_moments(R,options)
if ~exist('options','var')
options.show = 0;
end
if options.show == 1
disp('--- extracting moments...');
end
[Ireg,Jreg] = find(R==1); % pixels in the region
A = length(Ireg);
central = options.central;
rs = options.rs;
n = size(rs,1);
if central
i_m = mean(Ireg);
j_m = mean(Jreg);
I1 = Ireg - i_m*ones(A,1);
J1 = Jreg - j_m*ones(A,1);
strc = 'Central Moment';
else
I1 = Ireg;
J1 = Jreg;
strc = 'Moment';
end
X = zeros(1,n);
Xn = char(zeros(n,24));
for i=1:n
r = rs(i,1);
s = rs(i,2);
Ir = ones(A,1);
Js = ones(A,1);
if r>0
for jr = 1:r
Ir = Ir.*I1;
end
end
if s>0
for js = 1:s
Js = Js.*J1;
end
end
X(i) = Ir'*Js;
str = [strc ' ' num2str(r) ',' num2str(s) ' '];
Xn(i,:) = str(1:24);
end
|
github
|
domingomery/Balu-master
|
Bfx_basicint.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_basicint.m
| 1,843 |
utf_8
|
04c161a3907d7c57c57632eb0fc9accd
|
% [X,Xn] = Bfx_basicint(I,R,options)
% [X,Xn] = Bfx_basicint(I,options)
%
% Toolbox: Balu
% Basic intensity features
%
% X is the features vector, Xn is the list feature names (see Example to
% see how it works).
%
% Reference:
% Kumar, A.; Pang, G.K.H. (2002): Defect detection in textured materials
% using Gabor filters. IEEE Transactions on Industry Applications,
% 38(2):425-440.
%
% Example:
% options.mask = 5; % Gauss mask for gradient computation
% options.show = 1; % display results
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = double(I(:,:,2))/256; % normalized green channel
% [X,Xn] = Bfx_basicint(J,R,options); % basic intenisty features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_haralick, Bfx_clp, Bfx_gabor, Bfx_fourier, Bfx_dct, Bfx_lbp.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_basicint(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'mask')
options.mask = 15;
end
if options.show
disp('--- extracting basic intensity features...');
end
E = bwperim(R,4);
ii = find(R==1);
jj = find(R==0, 1);
kk = E==1;
I = double(I);
I1 = Bim_d1(I,options.mask);
I2 = Bim_d2(I);
if ~isempty(jj)
C = mean(abs(I1(kk)));
else
C = -1;
end
J = I(ii);
G = mean(J);
S = std(J);
K = kurtosis(J);
Sk = skewness(J);
D = mean(I2(ii));
X = [G S K Sk D C];
Xn = [ 'Intensity Mean '
'Intensity StdDev '
'Intensity Kurtosis '
'Intensity Skewness '
'Mean Laplacian '
'Mean Boundary Gradient '];
|
github
|
domingomery/Balu-master
|
Bfx_onesift.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_onesift.m
| 1,955 |
utf_8
|
c8fbe5fdc65eb5c2e40fbd368002f579
|
% [X,Xn,options] = Bfx_onesift(I,R,options)
% [X,Xn,options] = Bfx_onesift(I,options)
% [X,Xn] = Bfx_onesift(I,R,options)
% [X,Xn] = Bfx_onesift(I,options)
%
% Toolbox: Balu
% Extract only one SIFT descriptor of region R of image I.
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% R is a binary image or empty. If R is given the sift will be computed
% in the region defined by the piexels where R==1.
%
% Output:
%
% References:
% D. G. Lowe, Distinctive image features from scale-invariant
% keypoints. IJCV, vol. 2, no. 60, pp. 91-110, 2004.
%
% A. Vedaldi, B. Fulkerson: VLFeat: An Open and Portable Library
% of Computer Vision Algorithms, 2008 (http://www.vlfeat.org/)
%
%
% Example 5:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.25; % angle sampling
% options.weight = 9; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % weighted LBP features
% bar(X) % histogram
% See also Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_onesift(I,R,options)
%if nargin==2;
% options = R;
% R = ones(size(I));
%end
if isempty(R)
R = ones(size(I));
end
[ii,jj] = find(R==1);
ff = [mean(jj); mean(ii); sqrt(length(ii))/2; 0];
[ff,dd] = vl_sift(single(I),'Frames',ff);
X = dd';
n = 128;
Xn = char(zeros(n,24));
for k=1:n
s = sprintf('SIFT(%d) ',k);
Xn(k,:) = s(1:24);
end
|
github
|
domingomery/Balu-master
|
Bfx_hugeo.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_hugeo.m
| 2,304 |
utf_8
|
5e1a0fce78c514e4d097bc0f6941ba0e
|
% [X,Xn] = Bfx_hugeo(R)
% [X,Xn] = Bfx_hugeo(R,options)
%
% Toolbox: Balu
%
% Extract the seven Hu moments from binary image R.
%
% options.show = 1 display mesagges.
%
% X is a 7 elements vector:
% X(i): Hu-moment i for i=1,...,7.
% Xn is the list of feature names.
%
% Reference:
% Hu, M-K.: "Visual Pattern Recognition by Moment Invariants",
% IRE Trans. Info. Theory IT-8:179-187: 1962.
%
% Example:
% I = imread('testimg3.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [L,n] = bwlabel(R); % regions
% imshow(L,[])
% X = [];
% for i=1:n
% [Xi,Xn] = Bfx_hugeo(L==i); % Hu moments
% X = [X;Xi];
% end
% X
%
% See also Bfx_basicgeo, Bfx_gupta, Bfx_fitellipse, Bfx_flusser.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_hugeo(R,options)
if ~exist('options','var')
options.show = 0;
end
if options.show == 1
disp('--- extracting Hu moments...');
end
[Ireg,Jreg] = find(R==1); % pixels in the region
i_m = mean(Ireg);
j_m = mean(Jreg);
A = length(Ireg);
I0 = ones(A,1);
J0 = ones(A,1);
I1 = Ireg - i_m*ones(A,1);
J1 = Jreg - j_m*ones(A,1);
I2 = I1.*I1;
J2 = J1.*J1;
I3 = I2.*I1;
J3 = J2.*J1;
% Central moments
u00 = A; %u00 = m00 = (I0'*J0);
u002 = u00*u00;
u0025 = u00^2.5;
% u0015 = u00^1.5; not used
n02 = (I0'*J2)/u002;
n20 = (I2'*J0)/u002;
n11 = (I1'*J1)/u002;
n12 = (I1'*J2)/u0025;
n21 = (I2'*J1)/u0025;
n03 = (I0'*J3)/u0025;
n30 = (I3'*J0)/u0025;
f1 = n20+n02;
f2 = (n20-n02)^2 + 4*n11^2;
f3 = (n30-3*n12)^2+(3*n21-n03)^2;
f4 = (n30+n12)^2+(n21+n03)^2;
f5 = (n30-3*n12)*(n30+n12)*((n30+n12)^2 - 3*(n21+n03)^2) + (3*n21-n03)*(n21+n03)*(3*(n30+n12)^2 - (n21+n03)^2);
f6 = (n20-n02)*((n30+n12)^2 - (n21+n03)^2) + 4*n11*(n30+n12)*(n21+n03);
f7 = (3*n21-n03)*(n30+n12)*((n30+n12)^2 - 3*(n21+n03)^2) - (n30-3*n12)*(n21+n03)*(3*(n30+n12)^2 - (n21+n03)^2);
X = [f1 f2 f3 f4 f5 f6 f7];
Xn = [ 'Hu-moment 1 '
'Hu-moment 2 '
'Hu-moment 3 '
'Hu-moment 4 '
'Hu-moment 5 '
'Hu-moment 6 '
'Hu-moment 7 '];
|
github
|
domingomery/Balu-master
|
Bfx_hog.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_hog.m
| 3,258 |
utf_8
|
a089b1fa3b7f1cb82642615fc6cf1c1f
|
% [X,Xn] = Bfx_hog(I,options)
%
% Toolbox: Balu
% Histogram of Orientated Gradients features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% options.nj; : number of HOG windows per bound box
% options.ni : in i (vertical) and j (horizaontal) direction
% options.B : number of histogram bins
% options.show : show histograms (glyphs)
%
% Example:
% options.nj = 20; % 10 x 20
% options.ni = 10; % histograms
% options.B = 9; % 9 bins
% options.show = 1; % show results
% I = imread('testimg1.jpg'); % input image
% J = rgb2gray(I);
% figure(1);imshow(J,[]);
% figure(2);
% [X,Xn] = Bfx_hog(J,options); % HOG features (see gradients
% % arround perimeter).
%
% See also Bfx_phog, Bfx_lbp.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_hog(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'normalize')
options.normalize = 0;
end
nj = options.nj; % number of HOG windows per bound box
ni = options.ni; % in i (vertical) and j (horizaontal) direction
B = options.B; % number of histogram bins
show = options.show; % show histograms
N = size(I,1);
M = size(I,2);
X = zeros(1,nj*ni*B); % column vector with zeros
I = double(I);
dj = floor(M/(nj+1));
di = floor(N/(ni+1));
t = 0;
hj = [-1,0,1];
hi = -hj';
Gj = imfilter(I,hj);
Gi = imfilter(I,hi);
A = atan2(Gi,Gj);
A = mod(A,pi);
G = ((Gi.^2)+(Gj.^2)).^.5;
K = 4*di*dj;
ang = zeros(K,1);
mag = zeros(K,1);
if show
J = zeros(N,M);
ss = min([N/ni M/nj])*0.40;
end
w = zeros(ni,nj,B);
for i = 1:ni
ii = (i-1)*di+1:(i+1)*di;
i0 = mean(ii);
for j = 1:nj
jj = (j-1)*dj+1:(j+1)*dj;
j0 = mean(jj);
t = t+1;
ang(:) = A(ii,jj);
mag(:) = G(ii,jj);
X2 = zeros(B,1);
for b = 1:B
q = find(ang<=pi*b/B);
X2(b) = X2(b)+sum(mag(q));
ang(q) = 5;
end
X2 = X2/(norm(X2)+0.01);
X(1,indices(t,B)) = X2;
% w(i,j,:) = X2;
if show
for b=1:B
alpha = pi*b/B;
q = -ss:ss;
qi = round(i0+q*cos(alpha));
qj = round(j0+q*sin(alpha));
qk = qi+(qj-1)*N;
J(qk) = J(qk)+X2(b);
end
J(round(i0),round(j0))=1;
end
end
end
if show
imshow(round(J/max2(J)*256),jet)
options.J = J;
end
options.w = w;
Xn = char(zeros(nj*ni*B,24));
Xn(:,1) = 'H'; Xn(:,2) = 'O'; Xn(:,3) = 'G';
J = uint8(zeros(N,M,3));
G(G>363) = 363; % max2(Gi) = max2(Gi) = 256 => max2(G) = sqrt(2*256^2)
J(:,:,1) = uint8(round(G/363*255));
J(:,:,2) = uint8(round(A/pi*255));
options.Ihog = J;
if options.normalize
X = X/sum(X);
end
|
github
|
domingomery/Balu-master
|
Bfx_huint.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_huint.m
| 2,385 |
utf_8
|
2d95567f2dde7180bc05e62ba3fdb92d
|
% [X,Xn,Xu] = Bfx_huint(I,R,options)
%
% Toolbox: Balu
% Hu moments with intensity.
%
% X is a 7 elements vector:
% X(i): Hu-moment i for i=1,...,7.
% Xn is the list of feature names (see Example to see how it works).
%
% Reference:
% Hu, M-K.: "Visual Pattern Recognition by Moment Invariants",
% IRE Trans. Info. Theory IT-8:179-187: 1962.
%
% Example:
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = double(I(:,:,1))/256; % normalized red channel
% options.show = 1; % display results
% [X,Xn] = Bfx_huint(J,R,options); % Hu moments with intenisty
% Bio_printfeatures(X,Xn)
%
% See also Bfx_haralick, Bfx_clp, Bfx_fourier, Bfx_dct, Bfx_lbp.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_huint(I,R,options)
if ~exist('options','var')
options.show = 0;
end
if options.show == 1
disp('--- extracting Hu moments with intensity...');
end
[Ireg,Jreg] = find(R==1); % pixels in the region
im = mean(Ireg);
jm = mean(Jreg);
Kreg = R==1;
A = length(Ireg);
I0 = ones(A,1);
J0 = ones(A,1);
I1 = Ireg - im*ones(A,1);
J1 = Jreg - jm*ones(A,1);
I2 = I1.*I1;
J2 = J1.*J1;
I3 = I2.*I1;
J3 = J2.*J1;
xreg = I(Kreg);
if (sum(xreg==0))
xreg = ones(A,1);
end
J0X = double(J0).*double(xreg);
% Central moments
u00 = (I0'*J0X);
u002 = u00*u00;
u0025 = u00^2.5;
n02 = (I0'*J2)/u002;
n20 = (I2'*J0)/u002;
n11 = (I1'*J1)/u002;
n12 = (I1'*J2)/u0025;
n21 = (I2'*J1)/u0025;
n03 = (I0'*J3)/u0025;
n30 = (I3'*J0)/u0025;
f1 = n20+n02;
f2 = (n20-n02)^2 + 4*n11^2;
f3 = (n30-3*n12)^2+(3*n21-n03)^2;
f4 = (n30+n12)^2+(n21+n03)^2;
f5 = (n30-3*n12)*(n30+n12)*((n30+n12)^2 - 3*(n21+n03)^2) + (3*n21-n03)*(n21+n03)*(3*(n30+n12)^2 - (n21+n03)^2);
f6 = (n20-n02)*((n30+n12)^2 - (n21+n03)^2) + 4*n11*(n30+n12)*(n21+n03);
f7 = (3*n21-n03)*(n30+n12)*((n30+n12)^2 - 3*(n21+n03)^2) - (n30-3*n12)*(n21+n03)*(3*(n30+n12)^2 - (n21+n03)^2);
X = [f1 f2 f3 f4 f5 f6 f7];
Xn = [ 'Hu-moment-int 1 '
'Hu-moment-int 2 '
'Hu-moment-int 3 '
'Hu-moment-int 4 '
'Hu-moment-int 5 '
'Hu-moment-int 6 '
'Hu-moment-int 7 '];
|
github
|
domingomery/Balu-master
|
Bfx_lbpint.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbpint.m
| 18,156 |
utf_8
|
58b30c959a0a79e86ffcf6369760e789
|
% [X,Xn,options] = Bfx_lbp(I,R,options)
% [X,Xn,options] = Bfx_lbp(I,options)
% [X,Xn] = Bfx_lbp(I,R,options)
% [X,Xn] = Bfx_lbp(I,options)
%
% Toolbox: Balu
% Local Binary Patterns features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% It calculates the LBP over the a regular grid of patches. The function
% uses Heikkila & Ahonen (see http://www.cse.oulu.fi/MVG/Research/LBP).
%
% It returns a matrix of uniform lbp82 descriptors for I, made by
% concatenating histograms of each grid cell in the image.
% Grid size is options.hdiv * options.vdiv
%
% R is a binary image or empty. If R is given the lbp will be computed
% the corresponding pixles R==0 in image I will be set to 0.
%
% Output:
% X is a matrix of size ((hdiv*vdiv) x 59), each row has a
% histogram corresponding to a grid cell. We use 59 bins.
% options.x of size hdiv*vdiv is the x coordinates of center of ith grid cell
% options.y of size hdiv*vdiv is the y coordinates of center of ith grid cell
% Both coordinates are calculated as if image was a square of side length 1.
%
% References:
% Ojala, T.; Pietikainen, M. & Maenpaa, T. Multiresolution gray-scale
% and rotation invariant texture classification with local binary
% patterns. IEEE Transactions on Pattern Analysis and Machine
% Intelligence, 2002, 24, 971-987.
%
% Mu, Y. et al (2008): Discriminative Local Binary Patterns for Human
% Detection in Personal Album. CVPR-2008.
%
% Example 1:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'u2'; % uniform LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% bar(X) % histogram
%
% Example 2:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'ri'; % rotation-invariant LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% bar(X) % histogram
%
% Example 3:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 4:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 16; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 5:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.25; % angle sampling
% options.weight = 9; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % weighted LBP features
% bar(X) % histogram
% See also Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_lbpint(I,R,options)
if nargin==2;
options = R;
R = 1;
end
if ~isfield(options,'normalize')
options.normalize = 0;
end
if ~isfield(options,'coor')
i1 = 1;
j1 = 1;
i2 = size(I,1);
j2 = size(I,2);
else
i1 = options.coor(1);
j1 = options.coor(2);
i2 = options.coor(3);
j2 = options.coor(4);
end
desc = Bim_inthistread(I,i1,j1,i2,j2);
% vdiv = options.vdiv;
% hdiv = options.hdiv;
%
% if ~isfield(options,'show')
% options.show = 0;
% end
%
% if options.show == 1
% disp('--- extracting local binary patterns features...');
% end
%
% if ~isfield(options,'samples')
% options.samples = 8;
% end
%
% if ~isfield(options,'radius')
% options.radius = log(options.samples)/log(2)-1;
% end
%
% if ~isfield(options,'semantic')
% options.semantic = 0;
% end
%
% if ~isfield(options,'weight')
% options.weight = 0;
% end
LBPst = 'LBP';
st = 'int';
% if options.semantic>0
% if ~isfield(options,'sk')
% options.sk = 1;
% end
% mapping = getsmapping(options.samples,options.sk);
% LBPst = ['s' LBPst];
% st='8x8';
% else
% % mapping = getmapping(8,'u2');
% if ~isfield(options,'mappingtype')
% options.mappingtype = 'u2';
% end
% st = sprintf('%d,%s',options.samples,options.mappingtype);
% mapping = getmapping(options.samples,options.mappingtype);
% end
% get lbp image
% if ~isempty(R);
% I(R==0) = 0;
% end
code_img = I;
[n1,n2] = size(code_img);
% [N,M] = size(I);
% Ilbp = zeros(size(I));
% i1 = round((N-n1)/2);
% j1 = round((M-n2)/2);
% code_img = Ilbp(i1+1:i1+n1,j1+1:j1+n2);
% options.Ilbp = Ilbp;
%ylen = round(n1/vdiv);
%xlen = round(n2/hdiv);
% split image into blocks (saved as columns)
%grid_img = im2col(code_img,[ylen, xlen], 'distinct');
grid_img = code_img(:);
% if options.weight>0
% LBPst = ['w' LBPst];
% mt = 2*options.radius-1;
% mt2 = mt^2;
% Id = double(I);
% switch options.weight
% case 1
% W = abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id);
% case 2
% W = (abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id))./(Id+1);
% case 3
% W = abs(medfilt2(Id,[mt mt])-Id);
% case 4
% W = abs(medfilt2(Id,[mt mt])-Id)./(Id+1);
% case 5
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id);
% case 6
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
% case 7
% Id = conv2(Id,ones(mt,mt)/mt2,'same');
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
% case 8
% Id = medfilt2(Id,[mt mt]);
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
% case 9
% Id = medfilt2(Id,[mt mt]);
% W = abs(ordfilt2(Id,mt2-1,ones(mt,mt))-Id)./(Id+1);
% otherwise
% error('Bfx_lbp does not recognice options.weight = %d.',options.weight);
% end
% W = W(mt+1:end-mt,mt+1:end-mt);
% grid_W = im2col(W,[ylen, xlen], 'distinct');
% nwi = mapping.num;
% nwj = size(grid_W,2);
% nwk = size(grid_W,1);
% desc = zeros(nwi,nwj);
% for j=1:nwj
% x = grid_img(:,j)+1;
% y = grid_W(:,j);
% d = zeros(nwi,1);
% for k=1:nwk
% d(x(k))=d(x(k))+y(k);
% % d(x(k))=d(x(k))+1; % normal LBP each LBP has equal weight
% end
% desc(:,j) = d;
% end
%
% else
% desc = hist(double(grid_img), 0:options.maxD-1);
% calculate coordinates of descriptors as if I was square w/ side=1
%end
%dx = 1.0/hdiv;
%dy = 1.0/vdiv;
dx = 1;
dy = 1;
x = dx/2.0: dx :1.0-dx/2.0;
y = dy/2.0: dy :1.0-dy/2.0;
options.x = x;
options.y = y;
%if hdiv*vdiv>1
% D = desc';
%else
X = desc;
if options.normalize
X = X/sum(X);
end
%end
[M,N] = size(X);
Xn = char(zeros(N*M,24));
% X = zeros(1,N*M);
% k=0;
% for i=1:M
% for j=1:N
% k = k+1;
% s = sprintf('%s(%d,%d)[%s] ',LBPst,i,j,st);
% Xn(k,:) = s(1:24);
% X(k) = D(i,j);
% end
% end
end
%GETMAPPING returns a structure containing a mapping table for LBP codes.
% MAPPING = GETMAPPING(SAMPLES,MAPPINGTYPE) returns a
% structure containing a mapping table for
% LBP codes in a neighbourhood of SAMPLES sampling
% points. Possible values for MAPPINGTYPE are
% 'u2' for uniform LBP
% 'ri' for rotation-invariant LBP
% 'riu2' for uniform rotation-invariant LBP.
%
% Example:
% I=imread('rice.tif');
% MAPPING=getmapping(16,'riu2');
% LBPHIST=lbp(I,2,16,MAPPING,'hist');
% Now LBPHIST contains a rotation-invariant uniform LBP
% histogram in a (16,2) neighbourhood.
%
function mapping = getmapping(samples,mappingtype)
% Version 0.1.1
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% 0.1.1 Changed output to be a structure
% Fixed a bug causing out of memory errors when generating rotation
% invariant mappings with high number of sampling points.
% Lauge Sorensen is acknowledged for spotting this problem.
table = 0:2^samples-1;
newMax = 0; %number of patterns in the resulting LBP code
index = 0;
if strcmp(mappingtype,'u2') %Uniform 2
newMax = samples*(samples-1) + 3;
for i = 0:2^samples-1
j = bitset(bitshift(i,1,samples),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples)); %number of 1->0 and
%0->1 transitions
%in binary string
%x is equal to the
%number of 1-bits in
%XOR(x,Rotate left(x))
if numt <= 2
table(i+1) = index;
index = index + 1;
else
table(i+1) = newMax - 1;
end
end
end
if strcmp(mappingtype,'ri') %Rotation invariant
tmpMap = zeros(2^samples,1) - 1;
for i = 0:2^samples-1
rm = i;
r = i;
for j = 1:samples-1
r = bitset(bitshift(r,1,samples),1,bitget(r,samples)); %rotate
%left
if r < rm
rm = r;
end
end
if tmpMap(rm+1) < 0
tmpMap(rm+1) = newMax;
newMax = newMax + 1;
end
table(i+1) = tmpMap(rm+1);
end
end
if strcmp(mappingtype,'riu2') %Uniform & Rotation invariant
newMax = samples + 2;
for i = 0:2^samples - 1
j = bitset(bitshift(i,1,samples),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples));
if numt <= 2
table(i+1) = sum(bitget(i,1:samples));
else
table(i+1) = samples+1;
end
end
end
mapping.table=table;
mapping.samples=samples;
mapping.num=newMax;
end
% LBP returns the local binary pattern image or LBP histogram of an image.
% J = LBP(I,R,N,MAPPING,MODE) returns either a local binary pattern
% coded image or the local binary pattern histogram of an intensity
% image I. The LBP codes are computed using N sampling points on a
% circle of radius R and using mapping table defined by MAPPING.
% See the getmapping function for different mappings and use 0 for
% no mapping. Possible values for MODE are
% 'h' or 'hist' to get a histogram of LBP codes
% 'nh' to get a normalized histogram
% Otherwise an LBP code image is returned.
%
% J = LBP(I) returns the original (basic) LBP histogram of image I
%
% J = LBP(I,SP,MAPPING,MODE) computes the LBP codes using n sampling
% points defined in (n * 2) matrix SP. The sampling points should be
% defined around the origin (coordinates (0,0)).
%
% Examples
% --------
% I=imread('rice.png');
% mapping=getmapping(8,'u2');
% H1=LBP(I,1,8,mapping,'h'); %LBP histogram in (8,1) neighborhood
% %using uniform patterns
% subplot(2,1,1),stem(H1);
%
% H2=LBP(I);
% subplot(2,1,2),stem(H2);
%
% SP=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
% I2=LBP(I,SP,0,'i'); %LBP code image using sampling points in SP
% %and no mapping. Now H2 is equal to histogram
% %of I2.
function result = lbp(varargin) % image,radius,neighbors,mapping,mode)
% Version 0.3.2
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% Version 0.3.2: A bug fix to enable using mappings together with a
% predefined spoints array
% Version 0.3.1: Changed MAPPING input to be a struct containing the mapping
% table and the number of bins to make the function run faster with high number
% of sampling points. Lauge Sorensen is acknowledged for spotting this problem.
% Check number of input arguments.
error(nargchk(1,5,nargin));
image=varargin{1};
d_image=double(image);
if nargin==1
spoints=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
neighbors=8;
mapping=0;
mode='h';
end
if (nargin == 2) && (length(varargin{2}) == 1)
error('Input arguments');
end
if (nargin > 2) && (length(varargin{2}) == 1)
radius=varargin{2};
neighbors=varargin{3};
spoints=zeros(neighbors,2);
% Angle step.
a = 2*pi/neighbors;
for i = 1:neighbors
spoints(i,1) = -radius*sin((i-1)*a);
spoints(i,2) = radius*cos((i-1)*a);
end
if(nargin >= 4)
mapping=varargin{4};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 5)
mode=varargin{5};
else
mode='h';
end
end
if (nargin > 1) && (length(varargin{2}) > 1)
spoints=varargin{2};
neighbors=size(spoints,1);
if(nargin >= 3)
mapping=varargin{3};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 4)
mode=varargin{4};
else
mode='h';
end
end
% Determine the dimensions of the input image.
[ysize xsize] = size(image);
miny=min(spoints(:,1));
maxy=max(spoints(:,1));
minx=min(spoints(:,2));
maxx=max(spoints(:,2));
% Block size, each LBP code is computed within a block of size bsizey*bsizex
bsizey=ceil(max(maxy,0))-floor(min(miny,0))+1;
bsizex=ceil(max(maxx,0))-floor(min(minx,0))+1;
% Coordinates of origin (0,0) in the block
origy=1-floor(min(miny,0));
origx=1-floor(min(minx,0));
% Minimum allowed size for the input image depends
% on the radius of the used LBP operator.
if(xsize < bsizex || ysize < bsizey)
error('Too small input image. Should be at least (2*radius+1) x (2*radius+1)');
end
% Calculate dx and dy;
dx = xsize - bsizex;
dy = ysize - bsizey;
% Fill the center pixel matrix C.
C = image(origy:origy+dy,origx:origx+dx);
d_C = double(C);
bins = 2^neighbors;
% Initialize the result matrix with zeros.
result=zeros(dy+1,dx+1);
%Compute the LBP code image
for i = 1:neighbors
y = spoints(i,1)+origy;
x = spoints(i,2)+origx;
% Calculate floors, ceils and rounds for the x and y.
fy = floor(y); cy = ceil(y); ry = round(y);
fx = floor(x); cx = ceil(x); rx = round(x);
% Check if interpolation is needed.
if (abs(x - rx) < 1e-6) && (abs(y - ry) < 1e-6)
% Interpolation is not needed, use original datatypes
N = image(ry:ry+dy,rx:rx+dx);
D = N >= C;
else
% Interpolation needed, use double type images
ty = y - fy;
tx = x - fx;
% Calculate the interpolation weights.
w1 = (1 - tx) * (1 - ty);
w2 = tx * (1 - ty);
w3 = (1 - tx) * ty ;
w4 = tx * ty ;
% Compute interpolated pixel values
N = w1*d_image(fy:fy+dy,fx:fx+dx) + w2*d_image(fy:fy+dy,cx:cx+dx) + ...
w3*d_image(cy:cy+dy,fx:fx+dx) + w4*d_image(cy:cy+dy,cx:cx+dx);
D = N >= d_C;
end
% Update the result matrix.
v = 2^(i-1);
result = result + v*D;
end
%Apply mapping if it is defined
if isstruct(mapping)
bins = mapping.num;
for i = 1:size(result,1)
for j = 1:size(result,2)
result(i,j) = mapping.table(result(i,j)+1);
end
end
end
if (strcmp(mode,'h') || strcmp(mode,'hist') || strcmp(mode,'nh'))
% Return with LBP histogram if mode equals 'hist'.
result=hist(result(:),0:(bins-1));
if (strcmp(mode,'nh'))
result=result/sum(result);
end
else
%Otherwise return a matrix of unsigned integers
if ((bins-1)<=intmax('uint8'))
result=uint8(result);
elseif ((bins-1)<=intmax('uint16'))
result=uint16(result);
else
result=uint32(result);
end
end
end
% Mapping for sLBB
function mapping = getsmapping(N,sk)
M = 2^N;
samples = N;
len = zeros(M,1);
ang = zeros(M,1);
for x=0:(M-1)
k = x+1;
j = bitset(bitshift(x,1,samples),1,bitget(x,samples));
numt = sum(bitget(bitxor(x,j),1:samples));
c = numt;
if c>2
len(k)=-1;
ang(k)=-1;
else
s = bitget(x,1:samples);
len(k) = sum(s);
if c==0
ang(k)=0;
else
r = 0;
while (s(1)~=0) || (s(N)~=1)
s = [s(2:N) s(1)];
r = r+1;
end
ii = find(s==1);
a = mean(ii)+r;
if a>N
a=a-N;
end
ang(k) = round(sk*a)-1;
end
end
% fprintf('%4d: %s (%d,%d)\n',x,dec2bin(x,N),len(k),ang(k)); pause
end
Ma = max(ang)+1;
map = len*Ma+ang-Ma+1;
n = max(map)+1;
map(ang==-1) = n;
map(1) = 0;
mapping.table = map';
mapping.samples = N;
mapping.num = n+1;
end
|
github
|
domingomery/Balu-master
|
Bfx_basicgeo.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_basicgeo.m
| 2,763 |
utf_8
|
864db878ee7275da588d80683c9c45a7
|
% [X,Xn] = Bfx_geobasic(R,options)
%
% Toolbox: Balu
%
% Standard geometric features of a binary image R. This function calls
% regionprops of Image Processing Toolbox.
%
% options.show = 1 display mesagges.
%
% X is the feature vector
% Xn is the list of feature names.
%
% See expamle:
%
% Example:
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_basicgeo(R); % basic geometric features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_fitellipse, Bfx_hugeo, Bfx_gupta, Bfx_flusser.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_basicgeo(R,options)
if ~exist('options','var')
options.show = 0;
end
N = size(R,1);
warning off
stats = regionprops(uint8(R),'EulerNumber','ConvexArea','EquivDiameter','Solidity','MajorAxisLength','Extent','MinorAxisLength','Orientation','FilledArea','Eccentricity');
warning on
E = bwperim(R,4);
% center of gravity
[Ireg,Jreg] = find(R==1); % pixels in the region
Kreg = Ireg+Jreg*N-N; % pixels of region stored in a vector
i_m = mean(Ireg); % center of gravity in i direction
j_m = mean(Jreg); % center of gravity in i direction
% standard features
% Perimeter
if options.show == 1
disp('--- extracting standard geometric features...');
end
L8 = sum(sum(bwperim(R,8)));
L4 = sum(sum(bwperim(R,4)));
L = (3*L4+L8)/4;
% Area
A = bwarea(R);
% Roundness
Roundness = 4*A*pi/L^2;
% heigh & width
i_max = max(Ireg);
i_min = min(Ireg);
j_max = max(Jreg);
j_min = min(Jreg);
height = i_max-i_min+1; % height
width = j_max-j_min+1; % width
% Danielsson shape factor (see Danielsson, 1977)
TD = double(bwdist(not(R),'chessboard'));
dm = mean(TD(Kreg));
Gd = A/9/pi/dm^2;
X = [
i_m
j_m
height
width
A
L
Roundness
Gd
stats.EulerNumber
stats.EquivDiameter
stats.MajorAxisLength
stats.MinorAxisLength
stats.Orientation
stats.Solidity
stats.Extent
stats.Eccentricity
stats.ConvexArea
stats.FilledArea]';
Xn = [
'center of grav i [px] '
'center of grav j [px] '
'Height [px] '
'Width [px] '
'Area [px] '
'Perimeter [px] '
'Roundness '
'Danielsson factor '
'Euler Number '
'Equivalent Diameter [px]'
'MajorAxisLength [px] '
'MinorAxisLength [px] '
'Orientation [grad] '
'Solidity '
'Extent '
'Eccentricity '
'Convex Area [px] '
'Filled Area [px] '];
|
github
|
domingomery/Balu-master
|
Bfx_build.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_build.m
| 9,784 |
utf_8
|
b85a3456e0fcefa8ff8f3e6e8fa60df0
|
% bf = Bfx_build({'all'})
% bf = Bfx_build({'allgeo'})
% bf = Bfx_build({'allint'})
% bf = Bfx_build({'haralick','lbp'})
%
% Toolbox: Balu
% Build structure for feature extraction with default values
%
% Posible geometric names:
% 'basicgeo',
% 'fitellipse',
% 'fourierdes',
% 'hugeo',
% 'flusser',
% 'gupta',
%
% Posible intenisty names:
% 'basicint',
% 'contrast',
% 'haralick',
% 'lbp',
% 'lbps',
% 'dct',
% 'fourier',
% 'gabor',
% 'huint'
%
% Posible family names:
% 'all'
% 'allgeo'
% 'allint'
%
% Example:
% options.b = Bfx_build({'haralick','lbp'});
% options.colstr = 'rgb'; % R image
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_int(I,R,options); % intensity features
% Bio_printfeatures(X,Xn)
%
% See also Bcl_build, Bcl_balu.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function bfx = Bfx_build(varargin)
v = varargin;
n = nargin;
if compare(class(v),'cell')==0
v = v{1};
n = length(v);
end
if n==1
if compare(char(v(1)),'all')==0
v = {'basicgeo','fitellipse','fourierdes','hugeo','flusser','gupta','basicint','contrast','haralick','lbp','lbps','dct','fourier','gabor','huint'};
end
if compare(char(v(1)),'allgeo')==0
v = {'basicgeo','fitellipse','fourierdes','hugeo','flusser','gupta'};
end
if compare(char(v(1)),'allint')==0
v = {'basicint','contrast','haralick','lbp','lbps','dct','fourier','gabor','huint'};
end
end
n = length(v);
if n==0
bfx = [];
else
for k=1:n
s = char(v(k));
bfx(k).name = s;
switch lower(s)
% GEOMETRIC FEATURE EXTRACTION DEFINTION
case 'basicgeo'; % basic geometric features
bfx(k).options.type = 1; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'fitellipse'; % elliptic features
bfx(k).options.type = 1; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'fourierdes'; % Fourier descriptors
bfx(k).options.Nfourierdes = 16; % number of descriptors
bfx(k).options.type = 1; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'hugeo'; % Hu moments
bfx(k).options.type = 1; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'flusser'; % Flusser moments
bfx(k).options.type = 1; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'gupta'; % Gupta moments
bfx(k).options.type = 1; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
% INTENSITY FEATURE EXTRACTION DEFINTION
case 'basicint'; % basic intensity features
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'contrast'; % contrast features
bfx(k).options.neighbor = 2; % neigborhood is imdilate
bfx(k).options.param = 5; % with 5x5 mask
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'clp'; % crossing line profile (CLP)
bfx(k).options.ng = 32; % mask
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'haralick'; % statistical texture features
bfx(k).options.dharalick = 1:5; % for 1, 2, ... 5 pixels
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'lbp'; % local binary batterns (LBP)
bfx(k).options.vdiv = 1; % one vertical divition
bfx(k).options.hdiv = 1; % one horizontal divition
bfx(k).options.samples = 8; % number of neighbor samples
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'lbpri'; % local binary batterns (LBP)
bfx(k).name = 'lbp';
bfx(k).options.vdiv = 1; % one vertical divition
bfx(k).options.hdiv = 1; % one horizontal divition
bfx(k).options.samples = 8; % number of neighbor samples
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
bfx(k).options.mappingtype = 'ri'; % rotation invariant
case 'lbps'; % semantic LBP
bfx(k).name = 'lbp';
bfx(k).options.vdiv = 1; % one vertical divition
bfx(k).options.hdiv = 1; % one horizontal divition
bfx(k).options.semantic = 1; % semantic LBP
bfx(k).options.samples = 8; % number of neighbor samples
bfx(k).options.sk = 0.5; % angle sampling
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'lbpw'; % weighted semantic LBP
bfx(k).name = 'lbp';
bfx(k).options.vdiv = 1; % one vertical divition
bfx(k).options.hdiv = 1; % one horizontal divition
bfx(k).options.semantic = 0; % semantic LBP
bfx(k).options.weight = 9; % semantic LBP
bfx(k).options.samples = 8; % number of neighbor samples
bfx(k).options.sk = 0.5; % angle sampling
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case {'lbpws','lbpsw'}; % weighted & semantic LBP
bfx(k).name = 'lbp';
bfx(k).options.vdiv = 1; % one vertical divition
bfx(k).options.hdiv = 1; % one horizontal divition
bfx(k).options.semantic = 1; % semantic LBP
bfx(k).options.weight = 9; % semantic LBP
bfx(k).options.samples = 8; % number of neighbor samples
bfx(k).options.sk = 0.5; % angle sampling
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'dct'; % Discrete Cosinus Transform
bfx(k).options.Ndct = 64; % imresize vertical
bfx(k).options.Mdct = 64; % imresize horizontal
bfx(k).options.mdct = 4; % imresize frequency vertical
bfx(k).options.ndct = 4; % imresize frequency horizontal
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'fourier'; % Discrete Fourier Transform
bfx(k).options.Nfourier = 64; % imresize vertical
bfx(k).options.Mfourier = 64; % imresize horizontal
bfx(k).options.mfourier = 4; % imresize frequency vertical
bfx(k).options.nfourier = 4; % imresize frequency horizontal
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'gabor'; % Gabor features
bfx(k).options.Lgabor = 8; % number of rotations
bfx(k).options.Sgabor = 8; % number of dilations (scale)
bfx(k).options.fhgabor = 2; % highest frequency of interest
bfx(k).options.flgabor = 0.1; % lowest frequency of interest
bfx(k).options.Mgabor = 21; % mask size
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
case 'huint'; % Hu-moments with intensity
bfx(k).options.type = 2; % 1 geometric 2 intensity
bfx(k).options.show = 1; % display results
otherwise
error('Bfx_build does not recognize %s as feature extraction method.',s)
end
end
end
|
github
|
domingomery/Balu-master
|
Bfx_gupta.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_gupta.m
| 1,702 |
utf_8
|
9f533fbbc4a7c7b467d14e83acc0aa8c
|
% [X,Xn] = Bfx_gupta(R,options)
%
% Toolbox: Balu
%
% Extract the three Gupta moments from binary image R.
%
% options.show = 1 display mesagges.
%
% X is a 3 elements vector:
% X(i): Gupta-moment i for i=1,2,3.
% Xn is the list of feature names.
%
% Reference:
% Gupta, L. & Srinath, M. D. Contour sequence moments for the
% classification of closed planar shapes Pattern Recognition, 1987, 20,
% 267-272.
%
% Example:
% I = imread('testimg3.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [L,n] = bwlabel(R); % regions
% imshow(L,[])
% X = [];
% for i=1:n
% [Xi,Xn] = Bfx_gupta(L==i); % Gupta moments
% X = [X;Xi];
% end
% X
%
% See also Bfx_basicgeo, Bfx_hugeo, Bfx_fitellipse, Bfx_flusser.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_gupta(R,options)
if ~exist('options','var')
options.show = 0;
end
if options.show == 1
disp('--- extracting Gupta moments...');
end
E = bwperim(R,4);
[Ip,Jp] = find(E==1); % pixel of perimeter in (i,j)
jj = sqrt(-1);
Ig = Ip+jj*Jp;
ix = mean(Ip);
jx = mean(Jp);
centre = ix + jj*jx;
z = abs(Ig-centre);
m1 = mean(z);
mur1 = z-m1;
mur2 = mur1.*mur1;
mur3 = mur1.*mur2;
mur4 = mur2.*mur2;
mu2 = mean(mur2);
mu3 = mean(mur3);
mu4 = mean(mur4);
F1 = sqrt(mu2)/m1;
F2 = mu3/mu2/sqrt(mu2);
F3 = mu4/mu2^2;
X = [F1 F2 F3];
Xn = [ 'Gupta-moment 1 '
'Gupta-moment 2 '
'Gupta-moment 3 '];
|
github
|
domingomery/Balu-master
|
Bfx_clp.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_clp.m
| 4,246 |
utf_8
|
6693b6f561692d32557a46eb19e87152
|
% [X,Xn] = Bfx_clp(I,R,options)
% [X,Xn] = Bfx_clp(I,options)
%
% Toolbox Balu: Crossing Line Profile.
%
% X is the features vector, Xn is the list of feature names(see Example
% to see how it works).
%
% Reference:
% Mery, D.: Crossing line profile: a new approach to detecting defects
% in aluminium castings. Proceedings of the Scandinavian Conference on
% Image Analysis 2003 (SCIA 2003), Lecture Notes in Computer Science
% LNCS 2749: 725-732, 2003.
%
% Example:
% options.show = 1; % display results
% options.ng = 32; % windows resize
% I = imread('testimg4.jpg'); % input image
% J = I(395:425,415:442,1); % region of interest (red)
% R = J>135; % segmentation
% figure;imshow(J,[])
% figure;imshow(R)
% [X,Xn] = Bfx_clp(J,R,options); % CLP features
% Bio_printfeatures(X,Xn)
%
% See also Xcontrast, Xharalick, Bfx_clp, Xfourier, Xdct, Xlbp.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_clp(I,R,options)
[N,M] = size(I);
I = double(I);
if nargin==2;
options = R;
R = ones(N,M);
end
if ~isfield(options,'show')
options.show = 0;
end
if options.show == 1
disp('--- extracting Crossing line profile features...');
end
show = options.show;
[ii,jj] = find(R==1);
h = max(ii)-min(ii)+1; % height
w = max(jj)-min(jj)+1; % width
x = round(Xcentroid(R)); % mass center
i1 = max([1 x(1)-h]); j1 = max([1 x(2)-w]);
i2 = min([N x(1)+h]); j2 = min([M x(2)+w]);
%if ~isempty(R);
% I(R==0) = 0;
%end
ng = options.ng;
Bn = imresize(I(i1:i2,j1:j2),[ng ng]);
mg = fix(ng/2+1);
% Crossing line profiles
P0 = Bn(mg,:)'; % 0.0 ... 90.0 4
P1 = Bn(:,mg); % 90.0 ... 0.0 0
P2 = zeros(ng,1); % 45.0 ... 135.0 6
P3 = zeros(ng,1); % 135.0 ... 45.0 2
P4 = zeros(ng,1); % 22.5 ... 112.5 5
P5 = zeros(ng,1); % 67.5 ... 157.5 7
P6 = zeros(ng,1); % 112.5 ... 22.5 1
P7 = zeros(ng,1); % 157.5 ... 67.5 3
Q0 = Bn; Q1 = Bn; Q2 = Bn; Q3 = Bn;
Q4 = Bn; Q5 = Bn; Q6 = Bn; Q7 = Bn;
Q0(mg,:) = 255*ones(1,ng);
Q1(:,mg) = 255*ones(ng,1);
m4 = mg/(ng-1);
b4 = mg/2-m4;
b7 = 3*mg/2+mg/(ng-1);
for i=1:ng
P2(i,1) = Bn(i,i);
Q2(i,i) = 255;
P3(i,1) = Bn(i,ng-i+1);
Q3(i,ng-i+1) = 255;
j4 = fix(m4*i + b4 + 0.5);
P4(i,1) = Bn(j4,i);
Q4(j4,i) = 255;
P5(i,1) = Bn(i,j4);
Q5(i,j4) = 255;
j7 = fix(-m4*i + b7 + 0.5);
P6(i,1) = Bn(i,j7);
Q6(i,j7) = 255;
P7(i,1) = Bn(j7,i);
Q7(j7,i) = 255;
end
PP = [P0 P1 P2 P3 P4 P5 P6 P7];
d = abs(PP(1,:)-PP(ng,:));
[~,J] = sort(d);
Po = PP(:,J(1));
Po = Po/Po(1);
m = (Po(ng)-Po(1))/(ng - 1);
mb = Po(1)-m;
Q = Po-(1:ng)'*m-ones(ng,1)*mb;
Qm = mean(Q);
Qd = max(Q)-min(Q);
Qd1 = log(Qd+1);
Qd2 = 2*Qd/(Po(1)+Po(ng));
Qs = std(Q);
Qf = fft(Q);
Qf = abs(Qf(2:8,1));
if (show)
figure(10)
clf
subplot(2,4,5);plot(PP(:,1));axis([1 ng 0 255]);title('k=4');
subplot(2,4,1);plot(PP(:,2));axis([1 ng 0 255]);title('k=0');
subplot(2,4,3);plot(PP(:,3));axis([1 ng 0 255]);title('k=2');
subplot(2,4,7);plot(PP(:,4));axis([1 ng 0 255]);title('k=6');
subplot(2,4,4);plot(PP(:,5));axis([1 ng 0 255]);title('k=3');
subplot(2,4,2);plot(PP(:,6));axis([1 ng 0 255]);title('k=1');
subplot(2,4,8);plot(PP(:,7));axis([1 ng 0 255]);title('k=7');
subplot(2,4,6);plot(PP(:,8));axis([1 ng 0 255]);title('k=5');
figure(11)
imshow([Q1 Q5 Q2 Q4;Q0 Q7 Q3 Q6],gray(256));
pause(0);
end
X = [Qm Qs Qd Qd1 Qd2 Qf'];
Xn = [ 'CLP-Qm '
'CLP-Qs '
'CLP-Qd '
'CLP-Qd1 '
'CLP-Qd2 '
'CLP-Qf1 '
'CLP-Qf2 '
'CLP-Qf3 '
'CLP-Qf4 '
'CLP-Qf5 '
'CLP-Qf6 '
'CLP-Qf7 '];
|
github
|
domingomery/Balu-master
|
Bfx_gabor.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_gabor.m
| 3,540 |
utf_8
|
36d1e30d3fea9e840ac1b5566302615e
|
% [X,Xn] = Bfx_gabor(I,R,options)
% [X,Xn] = Bfx_gabor(I,options)
%
% Toolbox: Balu
% Gabor features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% Reference:
% Kumar, A.; Pang, G.K.H. (2002): Defect detection in textured materials
% using Gabor filters. IEEE Transactions on Industry Applications,
% 38(2):425-440.
%
% Example:
% options.Lgabor = 8; % number of rotations
% options.Sgabor = 8; % number of dilations (scale)
% options.fhgabor = 2; % highest frequency of interest
% options.flgabor = 0.1; % lowest frequency of interest
% options.Mgabor = 21; % mask size
% options.show = 1; % display results
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = I(:,:,2); % green channel
% [X,Xn] = Bfx_gabor(J,R,options); % Gabor features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_haralick, Bfx_clp, Bfx_fourier, Bfx_dct, Bfx_lbp.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_gabor(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
L = options.Lgabor;
S = options.Sgabor;
fh = options.fhgabor;
fl = options.flgabor;
M = options.Mgabor;
if options.show
disp('--- extracting Gabor features...');
end
alpha = (fh/fl)^(1/(S-1));
sx = sqrt(2*log(2))*(alpha+1)/2/pi/fh/(alpha-1);
sy = sqrt(2*log(2)-(2*log(2)/2/pi/sx/fh)^2)/(2*pi*tan(pi/2/L)*(fh-2*log(1/4/pi^2/sx^2/fh)));
u0 = fh;
k = R==1;
g = zeros(S,L);
size_out = size(I)+[M M]-1;
Iw = fft2(I,size_out(1),size_out(2));
n1 = (M+1)/2;
[NN,MM] = size(I);
for p=1:S;
for q=1:L
f = Bgabor_pq(p,q,L,sx,sy,u0,alpha,M);
%This convolution is very slow:
%Ir = conv2(I,real(f),'same');
%Ii = conv2(I,imag(f),'same');
%Iout = sqrt(Ir.*Ir + Ii.*Ii);
Ir = real(ifft2(Iw.*fft2(real(f),size_out(1),size_out(2))));
Ii = real(ifft2(Iw.*fft2(imag(f),size_out(1),size_out(2))));
Ir = Ir(n1:n1+NN-1,n1:n1+MM-1);
Ii = Ii(n1:n1+NN-1,n1:n1+MM-1);
Iout = sqrt(Ir.*Ir + Ii.*Ii);
g(p,q) = mean(Iout(k));
end;
end
gmax = max(g(:));
gmin = min(g(:));
J = (gmax-gmin)/gmin;
X = [g(:); gmax; gmin; J]';
LS = L*S;
Xn = char(zeros(LS+3,24));
k = 0;
for p=1:S;
for q=1:L
k = k + 1;
Xn(k,:) = sprintf('Gabor(%d,%d) ',p,q);
end
end
Xn(LS+1,:) = 'Gabor-max ';
Xn(LS+2,:) = 'Gabor-min ';
Xn(LS+3,:) = 'Gabor-J ';
% f = Bgabor_pq(p,q,L,S,sx,sy,u0,alpha,M)
%
% Toolbox: Balu
% Gabor kernel. See details in:
% Kumar, A.; Pang, G.K.H. (2002): Defect detection in textured materials
% using Gabor filters. IEEE Transactions on Industry Applications,
% 38(2):425-440.
%
% D.Mery, PUC-DCC, Apr. 2008
% http://dmery.ing.puc.cl
%
function f = Bgabor_pq(p,q,L,sx,sy,u0,alpha,M)
f = zeros(M,M);
sx2 = sx*sx;
sy2 = sy*sy;
c = (M+1)/2;
ap = alpha^-p;
tq = pi*(q-1)/L;
f_exp = 2*pi*sqrt(-1)*u0;
for i=1:M
x = i - c;
for j=1:M
y = j - c;
x1 = ap*(x*cos(tq)+y*sin(tq));
y1 = ap*(y*cos(tq)-x*sin(tq));
f(i,j) = exp(-0.5*(x1*x1/sx2+y1*y1/sy2))*exp(f_exp*x1);
end
end
f = ap*f/2/pi/sx/sy;
|
github
|
domingomery/Balu-master
|
Bfx_lbphogi.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbphogi.m
| 1,285 |
utf_8
|
697f23969b6f2ca4fe58208a9d934672
|
% [X,Xn] = Bfx_hog(I,options)
%
% Toolbox: Balu
% Histogram of Orientated Gradients features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% options.nj; : number of HOG windows per bound box
% options.ni : in i (vertical) and j (horizaontal) direction
% options.B : number of histogram bins
% options.show : show histograms (glyphs)
%
% Example:
% options.nj = 20; % 10 x 20
% options.ni = 10; % histograms
% options.B = 9; % 9 bins
% options.show = 1; % number of neighbor samples
% I = imread('testimg1.jpg'); % input image
% J = rgb2gray(I);
% figure(1);imshow(J,[]);
% figure(2);
% [X,Xn] = Bfx_hog(J,options); % HOG features (see gradients
% % arround perimeter.
%
% See also Bfx_phog, Bfx_lbp.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_lbphogi(I,R,options)
if nargin==2;
options = R;
R = [];
end
[X_lbp,Xn_lbp] = Bfx_lbpi(I,R,options);
[X_hog,Xn_hog] = Bfx_hogi(I,R,options);
X = [X_lbp X_hog ];
Xn = [Xn_lbp; Xn_hog];
|
github
|
domingomery/Balu-master
|
Bfx_gui.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_gui.m
| 47,257 |
utf_8
|
55d8165e26f39e53841b072ed7bd1721
|
% Bfx_gui
%
% Toolbox: Balu
%
% Graphic User Interface for feature extraction.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function varargout = Bfx_gui(varargin)
% BFX_GUI M-file for Bfx_gui.fig
% BFX_GUI, by itself, creates a new BFX_GUI or raises the existing
% singleton*.
%
% H = BFX_GUI returns the handle to a new BFX_GUI or the handle to
% the existing singleton*.
%
% BFX_GUI('CALLBACK',hObject,eventData,handles,...) calls the local
% function named CALLBACK in BFX_GUI.M with the given input arguments.
%
% BFX_GUI('Property','Value',...) creates a new BFX_GUI or raises the
% existing singleton*. Starting from the left, property value pairs are
% applied to the GUI before Bfx_gui_OpeningFcn gets called. An
% unrecognized property name or invalid value makes property application
% stop. All inputs are passed to Bfx_gui_OpeningFcn via varargin.
%
% *See GUI Options on GUIDE's Tools menu. Choose "GUI allows only one
% instance to run (singleton)".
%
% See also: GUIDE, GUIDATA, GUIHANDLES
% Edit the above text to modify the response to help Bfx_gui
% Last Modified by GUIDE v2.5 17-Nov-2011 10:21:48
% Begin initialization code - DO NOT EDIT
gui_Singleton = 1;
gui_State = struct('gui_Name', mfilename, ...
'gui_Singleton', gui_Singleton, ...
'gui_OpeningFcn', @Bfx_gui_OpeningFcn, ...
'gui_OutputFcn', @Bfx_gui_OutputFcn, ...
'gui_LayoutFcn', [] , ...
'gui_Callback', []);
if nargin && ischar(varargin{1})
gui_State.gui_Callback = str2func(varargin{1});
end
if nargout
[varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:});
else
gui_mainfcn(gui_State, varargin{:});
end
% End initialization code - DO NOT EDIT
% --- Executes just before Bfx_gui is made visible.
function Bfx_gui_OpeningFcn(hObject, eventdata, handles, varargin)
% This function has no output args, see OutputFcn.
% hObject handle to figure
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% varargin command line arguments to Bfx_gui (see VARARGIN)
% Choose default command line output for Bfx_gui
handles.output = hObject;
% Update handles structure
guidata(hObject, handles);
% UIWAIT makes Bfx_gui wait for user response (see UIRESUME)
% uiwait(handles.figure1);
clc
disp('Balu 3.0: GUI for Feature Extraction.')
disp(' ')
disp('This GUI will extract features from current directory:')
cd
disp('Warning: not all features implemented in Balu are in this GUI.');
disp(' ')
disp('Please define the feature extraction process in the GUI window,');
disp('and press [Go] when you finish.')
disp(' ')
global Img Imgnow fextractor Nimg sfiles pseg mseg seg fsname
fextractor.formatimg = 1;
fextractor.resizeimg = 1;
fextractor.geostandard = 0;
fextractor.geoinvariant = 0;
fextractor.geofourierdes = 0;
fextractor.geoelliptical = 0;
fextractor.intstandard = 0;
fextractor.intcontrast = 0;
fextractor.intharalick = 0;
fextractor.intfourierdct = 0;
fextractor.inthuint = 0;
fextractor.intgabor = 0;
fextractor.intlbp = 0;
fextractor.inthog = 0;
fextractor.colorgray = 0;
fextractor.colorred = 0;
fextractor.colorgreen = 0;
fextractor.colorblue = 0;
fextractor.colorhue = 0;
fextractor.colorsat = 0;
fextractor.colorval = 0;
fextractor.colorl = 0; %%%% Conversion RGB -> L*a*b*
fextractor.colora = 0; % 0 = no, 1 = calibrated (Bim_rgb2lab),
fextractor.colorb = 0; % 2 = formulas (Bim_rgb2lab0)
fextractor.segmentation = 0;
fextractor.partition = 1;
fextractor.outmatlab = 0;
fextractor.outascii = 0;
fextractor.outexcel = 0;
axis off
pseg = -0.05;
mseg = 0;
Nimg = 1;
seg = 0;
fsname = 0;
sfiles = dir('*.jpg');
if not(isempty(sfiles))
% error('Balu Error: Current directory does not contain any image')
%else
Img = imread(sfiles(Nimg).name);
Imgnow = Img;
subplot(1,1,1); imshow(Img)
title(sfiles(Nimg).name)
end
% --- Outputs from this function are returned to the command line.
function varargout = Bfx_gui_OutputFcn(hObject, eventdata, handles)
% varargout cell array for returning output args (see VARARGOUT);
% hObject handle to figure
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Get default command line output from handles structure
varargout{1} = handles.output;
% --- Executes on button press in ButtonGo.
function ButtonGo_Callback(hObject, eventdata, handles)
% hObject handle to ButtonGo (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
global fextractor
ok = 1;
fgeo = fextractor.geostandard+fextractor.geoinvariant+...
fextractor.geofourierdes+fextractor.geoelliptical;
fint = fextractor.intstandard+fextractor.intcontrast+...
fextractor.intharalick+fextractor.intfourierdct+...
fextractor.inthuint+fextractor.intgabor+fextractor.intlbp+...
fextractor.inthog;
fcol = fextractor.colorgray+fextractor.colorred+fextractor.colorgreen+...
fextractor.colorblue+fextractor.colorhue+fextractor.colorsat+...
fextractor.colorval+fextractor.colorl+fextractor.colora+...
fextractor.colorb;
fout = fextractor.outmatlab+fextractor.outascii+fextractor.outexcel;
if fgeo+fint==0
beep
wwarndlg('You must set some features.','Error in Bfx_gui');
ok = 0;
end
if and(fint>0,fcol==0)
beep
wwarndlg('If you select color features, you must set color component(s).','Error in Bfx_gui');
ok = 0;
end
if and(fint==0,fcol>0)
beep
wwarndlg('If you select color component(s), you must set color features.','Error in Bfx_gui');
ok = 0;
end
if (fout==0)
beep
wwarndlg('You must set output file format (Matlab, Text or Excel).','Error in Bfx_gui');
ok = 0;
end
if ok
yesnoans = questdlg('Bfx_gui will start to extract all features. This process could take several minutes. Are you sure?', ...
'Bfx_gui Information', ...
'Yes', 'No', 'No');
if yesnoans(1)=='Y'
save Bfx_guidata fextractor
[f,fn] = Bfx_guifun(fextractor);
questdlg(sprintf('Bfx_gui ended successfully: %d features extracted from %d images.',size(f,2),size(f,1)),'Bfx_gui Information','Ok', 'Ok');
end
end
% --- Executes on button press in Gray.
function Gray_Callback(hObject, eventdata, handles)
% hObject handle to Gray (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Gray
global Imgnow fextractor
val = get(hObject,'Value');
if val
if size(Imgnow,3)==1
subplot(1,1,1); imshow(Imgnow)
else
subplot(1,1,1); imshow(rgb2gray(Imgnow))
end
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorgray = val;
% --- Executes on button press in Red.
function Red_Callback(hObject, eventdata, handles)
% hObject handle to Red (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Red
global Imgnow fextractor
val = get(hObject,'Value');
if val
subplot(1,1,1); imshow(uint8(Imgnow(:,:,1)),[(1:256)'/256 zeros(256,2)])
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorred = val;
% --- Executes on button press in Green.
function Green_Callback(hObject, eventdata, handles)
% hObject handle to Green (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Green
global Imgnow fextractor
val = get(hObject,'Value');
if val
subplot(1,1,1); imshow(uint8(Imgnow(:,:,2)),[zeros(256,1) (1:256)'/256 zeros(256,1)])
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorgreen = val;
% --- Executes on button press in Blue.
function Blue_Callback(hObject, eventdata, handles)
% hObject handle to Blue (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Blue
global Imgnow fextractor
val = get(hObject,'Value');
if val
subplot(1,1,1); imshow(uint8(Imgnow(:,:,3)),[zeros(256,2) (1:256)'/256])
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorblue = val;
% --- Executes on button press in Hue.
function Hue_Callback(hObject, eventdata, handles)
% hObject handle to Hue (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Hue
global Imgnow fextractor
val = get(hObject,'Value');
if val
H = rgb2hsv(Imgnow);
H(:,:,2) = 1;
H(:,:,3) = 1;
subplot(1,1,1); imshow(hsv2rgb(H))
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorhue = val;
% --- Executes on button press in Sat.
function Sat_Callback(hObject, eventdata, handles)
% hObject handle to Sat (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Sat
global Imgnow fextractor
val = get(hObject,'Value');
if val
H = rgb2hsv(Imgnow);
subplot(1,1,1); imshow(H(:,:,2))
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorsat = val;
% --- Executes on button press in Val.
function Val_Callback(hObject, eventdata, handles)
% hObject handle to Val (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Val
global Imgnow fextractor
val = get(hObject,'Value');
if val
H = rgb2hsv(Imgnow);
subplot(1,1,1); imshow(H(:,:,3))
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorval = val;
% --- Executes on button press in L.
function L_Callback(hObject, eventdata, handles)
% hObject handle to L (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of L
global Imgnow fextractor
val = get(hObject,'Value');
if val
if (exist('LAB.mat','file'))
load LAB
L = Bim_rgb2lab(Imgnow,M);
subplot(1,1,1); imshow(L(:,:,1),[])
else
beep
yesnoans = questdlg('Calibrated LAB conversion cannot be possible because LAB.mat file does not exist. Do you want to convert RGB to L*a*b* using CIE formulas?', ...
'Bfx_gui RGB -> L*a*b conversion', ...
'Yes', 'No', 'No');
if yesnoans(1)=='Y'
val = 2;
L = Bim_rgb2lab0(Imgnow);
subplot(1,1,1); imshow(L(:,:,1),[])
else
wwarndlg('L*a*b* conversion is not possible, because LAB.mat file does not exist.','Error in Bfx_gui');
set(hObject,'Value',0);
val = 0;
end
end
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorl = val;
% --- Executes on button press in a.
function a_Callback(hObject, eventdata, handles)
% hObject handle to a (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of a
global Imgnow fextractor
val = get(hObject,'Value');
if val
if (exist('LAB.mat','file'))
load LAB
L = rgb2lab(Imgnow,M);
subplot(1,1,1); imshow(L(:,:,2),[])
else
beep
yesnoans = questdlg('Calibrated LAB conversion cannot be possible because LAB.mat file does not exist. Do you want to convert RGB to L*a*b* using CIE formulas?', ...
'Bfx_gui RGB -> L*a*b conversion', ...
'Yes', 'No', 'No');
if yesnoans(1)=='Y'
val = 2;
L = Bim_rgb2lab0(Imgnow);
subplot(1,1,1); imshow(L(:,:,2),[])
else
wwarndlg('L*a*b* conversion is not possible, because LAB.mat file does not exist.','Error in Bfx_gui');
set(hObject,'Value',0);
val = 0;
end
end
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colora = val;
% --- Executes on button press in b.
function b_Callback(hObject, eventdata, handles)
% hObject handle to b (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of b
global Imgnow fextractor
val = get(hObject,'Value');
if val
if (exist('LAB.mat','file'))
load LAB
L = rgb2lab(Imgnow,M);
subplot(1,1,1); imshow(L(:,:,3),[])
else
beep
yesnoans = questdlg('Calibrated LAB conversion cannot be possible because LAB.mat file does not exist. Do you want to convert RGB to L*a*b* using CIE formulas?', ...
'Bfx_gui RGB -> L*a*b conversion', ...
'Yes', 'No', 'No');
if yesnoans(1)=='Y'
val = 2;
L = Bim_rgb2lab0(Imgnow);
subplot(1,1,1); imshow(L(:,:,3),[])
else
wwarndlg('L*a*b* conversion is not possible, because LAB.mat file does not exist.','Error in Bfx_gui');
set(hObject,'Value',0);
val = 0;
end
end
else
subplot(1,1,1); imshow(Imgnow);
end
fextractor.colorb = val;
% --- Executes on button press in Matlab.
function Matlab_Callback(hObject, eventdata, handles)
% hObject handle to Matlab (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Matlab
global fextractor
fextractor.outmatlab = get(hObject,'Value');
% --- Executes on button press in Ascii.
function Ascii_Callback(hObject, eventdata, handles)
% hObject handle to Ascii (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Ascii
global fextractor
fextractor.outascii = get(hObject,'Value');
% --- Executes on button press in Excel.
function Excel_Callback(hObject, eventdata, handles)
% hObject handle to Excel (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Excel
global fextractor
fextractor.outexcel = get(hObject,'Value');
% --- Executes on button press in GeoStandard.
function GeoStandard_Callback(hObject, eventdata, handles)
% hObject handle to GeoStandard (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of GeoStandard
global fextractor
fextractor.geostandard = get(hObject,'Value');
if (fextractor.segmentation == 0) && (fextractor.geostandard == 1)
wwarndlg('Geometrical features with no segmentation? Bad idea :(','Warning in Bfx_gui');
end
% --- Executes on button press in IntMoments.
function IntMoments_Callback(hObject, eventdata, handles)
% hObject handle to IntMoments (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntMoments
global fextractor
fextractor.geoinvariant = get(hObject,'Value');
if (fextractor.segmentation == 0) && (fextractor.geoinvariant == 1)
wwarndlg('Geometrical features with no segmentation? Bad idea :(','Warning in Bfx_gui');
end
% --- Executes on button press in FourierDesc.
function FourierDesc_Callback(hObject, eventdata, handles)
% hObject handle to FourierDesc (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of FourierDesc
global fextractor
fextractor.geofourierdes = get(hObject,'Value');
if (fextractor.segmentation == 0) && (fextractor.geofourierdes == 1)
wwarndlg('Geometrical features with no segmentation? Bad idea :(','Warning in Bfx_gui');
end
% --- Executes on button press in Elliptical.
function Elliptical_Callback(hObject, eventdata, handles)
% hObject handle to Elliptical (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of Elliptical
global fextractor
fextractor.geoelliptical = get(hObject,'Value');
if (fextractor.segmentation == 0) && (fextractor.geoelliptical == 1)
wwarndlg('Geometrical features with no segmentation? Bad idea :(','Warning in Bfx_gui');
end
% --- Executes on button press in IntStandard.
function IntStandard_Callback(hObject, eventdata, handles)
% hObject handle to IntStandard (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntStandard
global fextractor
fextractor.intstandard = get(hObject,'Value');
% --- Executes on button press in IntContrast.
function IntContrast_Callback(hObject, eventdata, handles)
% hObject handle to IntContrast (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntContrast
global fextractor
fextractor.intcontrast = get(hObject,'Value');
% --- Executes on button press in IntHaralick.
function IntHaralick_Callback(hObject, eventdata, handles)
% hObject handle to IntHaralick (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntHaralick
global fextractor
fextractor.intharalick = get(hObject,'Value');
% --- Executes on button press in IntFourierDCT.
function IntFourierDCT_Callback(hObject, eventdata, handles)
% hObject handle to IntFourierDCT (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntFourierDCT
global fextractor
fextractor.intfourierdct = get(hObject,'Value');
% --- Executes on button press in IntHu.
function IntHu_Callback(hObject, eventdata, handles)
% hObject handle to IntHu (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntHu
global fextractor
fextractor.inthuint = get(hObject,'Value');
% --- Executes on button press in IntGabor.
function IntGabor_Callback(hObject, eventdata, handles)
% hObject handle to IntGabor (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntGabor
global fextractor
fextractor.intgabor = get(hObject,'Value');
% --- Executes on selection change in FormatImg.
function FormatImg_Callback(hObject, eventdata, handles)
% hObject handle to FormatImg (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hints: contents = get(hObject,'String') returns FormatImg contents as cell array
% contents{get(hObject,'Value')} returns selected item from FormatImg
global Img Imgnow fextractor Nimg sfiles
fextractor.formatimg = get(hObject,'Value');
s = dir('*.jpg');
if not(isempty(s))
% error('Balu Error: Current directory does not contain any image')
%else
end
switch fextractor.formatimg
case 2
s = 'tif';
case 3
s = 'bmp';
case 4
s = 'png';
case 5
s = 'ppm';
case 6
s = 'gif';
case 7
s = 'pbm';
otherwise
s = 'jpg';
end
if isempty(s)
beep
wwarndlg('Current directory does not contain any image with this format.','Error in Bfx_gui');
else
sfiles = [dir(['*.' lower(s)]); dir(['*.' upper(s)])];
Nimg = 1;
Img = imread(sfiles(1).name);
Imgnow = Img;
subplot(1,1,1); imshow(Img)
title(sfiles(Nimg).name)
end
% --- Executes during object creation, after setting all properties.
function FormatImg_CreateFcn(hObject, eventdata, handles)
% hObject handle to FormatImg (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles empty - handles not created until after all CreateFcns called
% Hint: popupmenu controls usually have a white background on Windows.
% See ISPC and COMPUTER.
if ispc && isequal(get(hObject,'BackgroundColor'), get(0,'defaultUicontrolBackgroundColor'))
set(hObject,'BackgroundColor','white');
end
function ImgResize_Callback(hObject, eventdata, handles)
% hObject handle to ImgResize (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hints: get(hObject,'String') returns contents of ImgResize as text
% str2double(get(hObject,'String')) returns contents of ImgResize as a double
global fextractor Img Imgnow
fextractor.resizeimg = str2num(get(hObject,'String'));
warning off
Imgnow = imresize(Img,fextractor.resizeimg);
warning on
subplot(1,1,1); imshow(Imgnow)
% --- Executes during object creation, after setting all properties.
function ImgResize_CreateFcn(hObject, eventdata, handles)
% hObject handle to ImgResize (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles empty - handles not created until after all CreateFcns called
% Hint: edit controls usually have a white background on Windows.
% See ISPC and COMPUTER.
if ispc && isequal(get(hObject,'BackgroundColor'), get(0,'defaultUicontrolBackgroundColor'))
set(hObject,'BackgroundColor','white');
end
% --- Executes on selection change in ImgSegmentation.
function ImgSegmentation_Callback(hObject, eventdata, handles)
% hObject handle to ImgSegmentation (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hints: contents = get(hObject,'String') returns ImgSegmentation contents as cell array
% contents{get(hObject,'Value')} returns selected item from ImgSegmentation
global fextractor Imgnow pseg seg mseg fsname
fextractor.segmentation = get(hObject,'Value')-1;
seg = fextractor.segmentation;
fsname = 0;
PlotImage(Imgnow,seg,pseg,mseg);
function [R,E,J] = SegImage(Imgnow,seg,pseg,mseg)
global fsname
[N,M,P] = size(Imgnow);
switch seg
case 1 % balu
[R,E,J] = Bim_segbalu(Imgnow,pseg);
case 2 % maxvar
[R,E,J] = Bim_segmaxvar(Imgnow,pseg);
case 3 % maxfisher
[R,E,J] = Bim_segmaxfisher(Imgnow,pseg);
case 4 % otsu
[R,E,J] = Bim_segotsu(Imgnow,pseg);
case 5
[R,E,J] = Bim_segpca(Imgnow,pseg);
case 6
if fsname == 0
sname = input('Input segmentation program name: ');
fsname = ['[R,E,J] = ' sname '(Imgnow,pseg);'];
end
eval(fsname);
otherwise
R = ones(N,M);
if (P==3)
J = rgb2gray(Imgnow);
else
J = Imgnow;
end
R = ones(size(J));
E = zeros(size(J));
end
if mseg
R = bwlabel(R);
end
function PlotImage(Imgnow,seg,pseg,mseg)
[R,E,J] = SegImage(Imgnow,seg,pseg,mseg);
[N,M,P] = size(Imgnow);
subplot(1,1,1); imshow(R,[])
pause(1)
ii = find(R==0);
if isempty('RR')
Iw = Imgnow;
else
if (P==3)
Ir = Imgnow(:,:,1);Ir(ii)=0;
Ig = Imgnow(:,:,2);Ig(ii)=0;
Ib = Imgnow(:,:,3);Ib(ii)=0;
Iw = zeros(size(Imgnow));
Iw(:,:,1) = Ir;
Iw(:,:,2) = Ig;
Iw(:,:,3) = Ib;
else
Iw = Imgnow;
Iw(ii) = 0;
end
subplot(1,1,1); imshow(uint8(Iw))
end
% --- Executes during object creation, after setting all properties.
function ImgSegmentation_CreateFcn(hObject, eventdata, handles)
% hObject handle to ImgSegmentation (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles empty - handles not created until after all CreateFcns called
% Hint: popupmenu controls usually have a white background on Windows.
% See ISPC and COMPUTER.
if ispc && isequal(get(hObject,'BackgroundColor'), get(0,'defaultUicontrolBackgroundColor'))
set(hObject,'BackgroundColor','white');
end
% --- Executes on button press in lbp.
function lbp_Callback(hObject, eventdata, handles)
% hObject handle to lbp (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of lbp
global fextractor
fextractor.intlbp = get(hObject,'Value');
% --- Executes on selection change in Partition.
function Partition_Callback(hObject, eventdata, handles)
% hObject handle to Partition (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hints: contents = get(hObject,'String') returns Partition contents as cell array
% contents{get(hObject,'Value')} returns selected item from Partition
global fextractor Imgnow
fextractor.partition = get(hObject,'Value');
[N,M,P] = size(Imgnow);
n = N/fextractor.partition;
m = M/fextractor.partition;
subplot(1,1,1); imshow(Imgnow)
hold on
for i=0:n:N
plot([1 M],[i i])
end
for j=0:m:M
plot([j j],[1 N])
end
hold off
% --- Executes during object creation, after setting all properties.
function Partition_CreateFcn(hObject, eventdata, handles)
% hObject handle to Partition (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles empty - handles not created until after all CreateFcns called
% Hint: popupmenu controls usually have a white background on Windows.
% See ISPC and COMPUTER.
if ispc && isequal(get(hObject,'BackgroundColor'), get(0,'defaultUicontrolBackgroundColor'))
set(hObject,'BackgroundColor','white');
end
% Bfx_guifun
%
% Toolbox: Balu
% Extract feactures of all images of courrent directory and store.
% The extracted features are stored in excel, text or matlab files.
% A list of the names of the images is stored in imgnames.txt
% f = table of features, fn = name of features
%
% fetractor is a structure with following variables:
%
% fextractor.formatimg 1=jpg|2=tif|3=bmp|4=png|5=ppm|6=png|7=pbm
% fextractor.resizeimg 0=no|0.5, 0.25, [16 16], [32 32], [200 200]
% fextractor.geostandard 1=yes|0=no
% fextractor.geoinvariant :
% fextractor.geofourierdes
% fextractor.geoelliptical
% fextractor.intstandard
% fextractor.intcontrast :
% fextractor.intharalick
% fextractor.intfourierdct
% fextractor.inthuint
% fextractor.intgabor
% fextractor.intlbp
% fextractor.colorgray
% fextractor.colorred :
% fextractor.colorgreen
% fextractor.colorblue
% fextractor.colorhue
% fextractor.colorsat
% fextractor.colorval :
% fextractor.colorl
% fextractor.colora
% fextractor.colorb
% fextractor.segmentation 1=yes | 0 = no (whole image)
% fextractor.outmatlab
% fextractor.outascii
% fextractor.outexcel
%
% D.Mery, PUC-DCC, Apr. 2008
% http://dmery.ing.puc.cl
%
function [f,fn]=Bfx_guifun(fextractor)
fsb = Bio_statusbar('Extracting features');
global pseg mseg seg
disp('Balu Full Features Extractor');
disp(' ');
reset(gca);
imgfmt = fextractor.formatimg;
fg = zeros(10,1);
fi = zeros(10,1);
co = zeros(20,1);
fg(1) = fextractor.geostandard;
fg(2) = fextractor.geoinvariant;
fg(3) = fextractor.geofourierdes;
fg(4) = fextractor.geoelliptical;
rg = sum(fg)>0;
fi(1) = fextractor.intstandard;
fi(2) = fextractor.intcontrast;
fi(3) = fextractor.intharalick;
fi(4) = fextractor.intfourierdct;
fi(5) = fextractor.inthuint;
fi(6) = fextractor.intgabor;
fi(7) = fextractor.intlbp;
fi(8) = fextractor.inthog;
ri = sum(fi)>0;
co(1) = fextractor.colorgray;
co(2) = fextractor.colorred;
co(3) = fextractor.colorgreen;
co(4) = fextractor.colorblue;
co(5) = fextractor.colorhue;
co(6) = fextractor.colorsat;
co(7) = fextractor.colorval;
co(8) = fextractor.colorl;
co(9) = fextractor.colora;
co(10)= fextractor.colorb;
if and(sum(co)>0,sum(fi)==0)
disp('Warning: No intensity feature will be extracted, because no intensity feature was marked. ')
ri = 0;
end
if and(sum(co)==0,sum(fi)>0)
disp('Warning: No intensity feature will be extracted, because no color component was marked. ')
ri = 0;
end
if (sum(co(8:10)>0)>0)
rgblabconv = (sum(co(8:10)==2)>0)+1;
else
rgblabconv = 0;
end
if (sum([fg;fi])==0)
error('No feature will be extracted, because there is no feature marked. ')
end
imgseg = fextractor.segmentation;
imgpart = fextractor.partition;
if imgpart == 0
imgpart = 1; % no partition means one partition!
end
oxls = fextractor.outexcel;
otxt = fextractor.outascii;
omat = fextractor.outmatlab;
imgres = fextractor.resizeimg;
if rg
rgi = 0;
% fg(1) = fextractor.geostandard;
if fg(1)
rgi = rgi+1;
bg(rgi).name = 'basicgeo';
bg(rgi).options.show = 0;
end
% fg(2) = fextractor.geoinvariant;
if fg(2)
rgi = rgi+1;
bg(rgi).name = 'hugeo';
bg(rgi).options.show = 0;
rgi = rgi+1;
bg(rgi).name = 'flusser';
bg(rgi).options.show = 0;
end
% fg(3) = fextractor.geofourierdes;
if fg(3)
rgi = rgi+1;
bg(rgi).name = 'fourierdes';
bg(rgi).options.show = 0;
bg(rgi).options.Nfourierdes = 8;
end
% fg(4) = fextractor.geoelliptical;
if fg(4)
rgi = rgi+1;
bg(rgi).name = 'fitellipse';
bg(rgi).options.show = 0;
end
opg.b = bg;
end
if ri
rii = 0;
% fi(1) = fextractor.intstandard;
if fi(1)
rii = rii+1;
bi(rii).name = 'basicint';
bi(rii).options.show = 0;
end
% fi(2) = fextractor.intcontrast;
if fi(2)
rii = rii+1;
bi(rii).name = 'contrast';
bi(rii).options.neighbor = 1;
bi(rii).options.param = 1.5;
bi(rii).options.show = 0;
end
% fi(3) = fextractor.intharalick;
if fi(3)
rii = rii+1;
bi(rii).name = 'haralick';
bi(rii).options.dharalick = 1;
bi(rii).options.show = 0;
rii = rii+1;
bi(rii).name = 'haralick';
bi(rii).options.dharalick = 2;
bi(rii).options.show = 0;
rii = rii+1;
bi(rii).name = 'haralick';
bi(rii).options.dharalick = 3;
bi(rii).options.show = 0;
rii = rii+1;
bi(rii).name = 'haralick';
bi(rii).options.dharalick = 4;
bi(rii).options.show = 0;
rii = rii+1;
bi(rii).name = 'haralick';
bi(rii).options.dharalick = 5;
bi(rii).options.show = 0;
end
% fi(4) = fextractor.intfourierdct;
if fi(4)
rii = rii+1;
bi(rii).name = 'fourier';
bi(rii).options.Nfourier = 64;
bi(rii).options.Mfourier = 64;
bi(rii).options.nfourier = 4;
bi(rii).options.mfourier = 4;
bi(rii).options.show = 0;
rii = rii+1;
bi(rii).name = 'dct';
bi(rii).options.Ndct = 64;
bi(rii).options.Mdct = 64;
bi(rii).options.ndct = 4;
bi(rii).options.mdct = 4;
bi(rii).options.show = 0;
end
% fi(5) = fextractor.inthuint;
if fi(5)
rii = rii+1;
bi(rii).name = 'huint';
bi(rii).options.show = 0;
end
% fi(6) = fextractor.intgabor;
if fi(6)
rii = rii+1;
bi(rii).name = 'gabor';
bi(rii).options.Lgabor = 8;
bi(rii).options.Sgabor = 8;
bi(rii).options.fhgabor = 2;
bi(rii).options.flgabor = 0.1;
bi(rii).options.Mgabor = 21;
bi(rii).options.show = 0;
end
% fi(7) = fextractor.intlbp;
if fi(7)
rii = rii+1;
bi(rii).name = 'lbp';
bi(rii).options.vdiv = 1;
bi(rii).options.hdiv = 1;
bi(rii).options.semantic = 0;
bi(rii).options.samples = 8;
bi(rii).options.mappingtype = 'u2';
bi(rii).options.show = 0;
end
if fi(8)
rii = rii+1;
bi(rii).name = 'hog';
bi(rii).options.ni = 1;
bi(rii).options.nj = 1;
bi(rii).options.B = 9;
bi(rii).options.show = 0;
end
opi.b = bi;
end
warning off
if ispc
switch imgfmt
case 2
!dir *.tif/B > imgnames.txt
sdir = dir('*.tif');
case 3
!dir *.bmp/B > imgnames.txt
sdir = dir('*.bmp');
case 4
!dir *.png/B > imgnames.txt
sdir = dir('*.png');
case 5
!dir *.ppm/B > imgnames.txt
sdir = dir('*.ppm');
case 6
!dir *.gif/B > imgnames.txt
sdir = dir('*.gif');
case 7
!dir *.pbm/B > imgnames.txt
sdir = dir('*.pbm');
otherwise
!dir *.jpg/B > imgnames.txt
sdir = dir('*.jpg');
end
else % Asuming unix, ie isunix should be 1
switch imgfmt
case 2
!ls *.tif > imgnames.txt
sdir = dir('*.tif');
case 3
!ls *.bmp > imgnames.txt
sdir = dir('*.bmp');
case 4
!ls *.png > imgnames.txt
sdir = dir('*.png');
case 5
!ls *.ppm > imgnames.txt
sdir = dir('*.ppm');
case 6
!ls *.png > imgnames.txt
sdir = dir('*.png');
case 7
!ls *.pbm > imgnames.txt
sdir = dir('*.pbm');
otherwise
!ls *.jpg > imgnames.txt
sdir = dir('*.jpg');
end
end
fid = fopen('imgnames.txt','rt');
sdirn = length(sdir);
ok = 1;
if rg
FeatureGeo = [];
end
if ri
neigh = 0;
%costr = ['Gray '; 'Red '; 'Green'; 'Blue '; 'Hue '; 'Sat '; 'Value'; 'L '; 'a '; 'b ' ];
costr = 'gRGBHSVLab';
color_str = costr(co==1);
%for i=1:10
% if co(i)
% s = sprintf('Feature%s = [];',costr(i,:));
% eval(s);
% end
%end
end
fall = []; % extracted features
falln = []; % feature names
imgnames = [];
sdiri = 0;
while(ok)
fsb = Bio_statusbar(sdiri/sdirn,fsb);
sdiri = sdiri+1;
s = fscanf(fid,'%s\n',[1 1]);
if isempty(s)
ok = 0;
else
disp(sprintf('\n\nProcessing image %s...',s));
Io = imread(s);
ss = [s ' '];
sname = ss(1:32);
%[N,M,P] = size(Io);
Io = double(Io);
if imgres(1)>0
% k = 2^(imgres+3);
% I = imresize(Io,[k k]);
mini = min(Io(:));
maxi = max(Io(:));
I = Bim_sat(imresize(Io,imgres),mini,maxi);
else
I = Io;
end
[N,M,P] = size(I);
% segmentation yes/no
[R,E,J] = SegImage(I,seg,pseg,mseg);
if sum2(R)==0
if P==3 % 3 channels
subplot(2,2,1);imshow(I/256,[]); title(s)
else
subplot(2,2,1);imshow(Io,[]); title(s)
end
subplot(2,2,2);imshow(R,[]);title('Segmented Region')
str = ['Image ' s ' has no segmentation. The feature extraction will fail. Do you want to segment the whole image?'];
yesnoans = questdlg(str, ...
'Bfx_gui: Segmentation Failure', ...
'Yes', 'Abort', 'Abort');
if yesnoans(1)=='Y'
R = ones(N,M);
end
end
% switch imgseg
% case 1
% [R,E,J] = Bim_segbalu(I);
% case 2
% [R,E,J] = Bim_segbalu(I,0.20);
% case 3
% [R,E,J] = Bim_segbalu(I,-0.20);
% case 4
% [R,E,J] = Bim_segbalu(I);
% R = bwlabel(R);
% case 5
% [R,E,J] = Bim_segbalu(I,0.20);
% R = bwlabel(R);
% case 6
% [R,E,J] = Bim_segbalu(I,-0.20);
% R = bwlabel(R);
% otherwise
% R = ones(N,M);
% if (P==3)
% J = rgb2gray(I);
% else
% J = I;
% end
% end
% I: imagen a procesar (color or grayvalues)
% J: imagen en gray values (transformed or original)
% R: segmented image (if no R = ones(N,M)
% color vs. grayvalues
if (P==3)
Imgnow = uint8(round(I));
else
Imgnow = I;
end
ii = find(R==0);
if P==3 % 3 channels
subplot(2,2,1);imshow(I/256,[]); title(s)
Ir = Imgnow(:,:,1);
Ig = Imgnow(:,:,2);
Ib = Imgnow(:,:,3);
else
subplot(2,2,1);imshow(Io,[]); title(s)
Iw = Imgnow;
end
if not(isempty(ii))
if (P==3)
Ir(ii)=0;
Ig(ii)=0;
Ib(ii)=0;
else
Iw(ii) = 0;
end
end
if P==3
Iw = zeros(size(Imgnow));
Iw(:,:,1) = Ir;
Iw(:,:,2) = Ig;
Iw(:,:,3) = Ib;
imshow(uint8(Iw))
else
imshow(Iw,[])
end
title(s)
subplot(2,2,2);imshow(R,[]);title('Segmented Region')
pause(0.1)
RRR = R;
III = I;
dpi = fix(N/imgpart);
dpj = fix(M/imgpart);
fpart = [];
fnpart = [];
fnimg = []; % feature names
fimg = []; % fetaures for this image
for parti=1:imgpart
for partj=1:imgpart
if imgpart > 1
spart = [num2str(parti) num2str(partj)];
else
spart = ' ';
end
fn = []; % feature names
fs = []; % extracted features for this partition
R = RRR(dpi*(parti-1)+1:dpi*parti,dpj*(partj-1)+1:dpj*partj);
I = III(dpi*(parti-1)+1:dpi*parti,dpj*(partj-1)+1:dpj*partj,:);
%figure(10)
%imshow(I/256,[])
%figure(11)
%imshow(III/256,[])
if rg
%[NameGeo,Feature,UnitGeo] = geofeatures(R,fg);
[Feature,NameGeo] = Bfx_geo(R,opg);
%FeatureGeo = [FeatureGeo; Feature];
fn = [fn; NameGeo];
fs = [fs Feature];
end
if ri
if sum(co(2:4))>0
XRGB = I;
end
if sum(co(5:7))>0
XHSV = rgb2hsv(I);
end
if rgblabconv>0
if rgblabconv==1
load LAB
XLAB = Bim_rgb2lab(I,M);
else
XLAB = Bim_rgb2lab0(I);
end
end
color_str = '';
for i=1:10
if co(i)
switch i
case 1 % Gray
if size(I,3)==3
X = rgb2gray(I/256)*256;
Xh = uint8(round(X));
else
X = double(I);
if max(X(:))>256
Xh = uint8(double(X)/256);
X = X/256;
else
Xh = uint8(X);
end
end
case 2 % Red
X = XRGB(:,:,1);
Xh = uint8(X);
case 3 % Green
X = XRGB(:,:,2);
Xh = uint8(X);
case 4 % Blue
X = XRGB(:,:,3);
Xh = uint8(X);
case 5 % H
X = XHSV(:,:,1);
Xh = double(X);
case 6 % S
X = XHSV(:,:,2);
Xh = double(X);
case 7 % V
X = XHSV(:,:,3);
Xh = uint8(X);
case 8 % L*
X = XLAB(:,:,1);
Xh = double(X)/100;
case 9 % a*
X = XLAB(:,:,2);
Xh = (double(X)+120)/240;
case 10 % b*
X = XLAB(:,:,3);
Xh = (double(X)+120)/240;
end
opi.colstr = costr(i);
[Feature,FeatureName] = Bfx_int(X,R,opi);
fn = [fn; FeatureName];
fs = [fs Feature];
subplot(2,2,3);imshow(X,[]);title(['Channel-' costr(i)])
%if (P==3)
% subplot(2,2,4);imhist(X);title([costr(i,:) ' Histogram'])
%else
subplot(2,2,4);imhist(Xh);title(['Histogram-' costr(i)])
%end
pause(0)
end
end
end
fn = [fn ones(size(fn,1),1)*spart];
fnimg = [fnimg;fn]; % feature names
fimg = [fimg fs]; % fetaures for this image
for si = 1:size(fs,1)
imgnames = [imgnames; sname];
end
end
end
fall = [fall; fimg];
end
fnall = fnimg;
save % ..backup
end
f = fall;
fn = fnall;
fclose(fid);
if omat
disp('Bfx_gui: saving f (features) and fn (feature names) to Bfx_results.mat...')
save Bfx_results f fn imgnames
end
if oxls
% balu2xls
disp('Bfx_gui: saving features to Bfx_results.csv (excel compatible)...')
csvwrite('Bfx_results.csv',f);
disp('Saving feature names in Bfx_guinames.txt...')
save Bfx_imgnames.txt fn -ascii
end
if otxt
disp('Bfx_gui: saving features to Bfx_results.txt...')
save Bfx_results.txt f -ascii -double
disp('Saving feature names in Bfx_guinames.txt...')
save Bfx_imgnames.txt fn -ascii
end
warning on
disp(' ')
disp(sprintf('Bfx_gui ended successfully: %d features extracted from %d images.',size(f,2),size(f,1)))
disp('(all variables saved in matlab.mat as backup)')
delete(fsb);
% --- Executes on button press in PreviousImage.
function PreviousImage_Callback(hObject, eventdata, handles)
% hObject handle to PreviousImage (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
global Nimg sfiles Imgnow
nf = length(sfiles);
if nf>0
Nimg = Nimg-1;
if Nimg<1
Nimg = nf;
end
Img = imread(sfiles(Nimg).name);
Imgnow = Img;
subplot(1,1,1); imshow(Img)
title(sfiles(Nimg).name)
end
% --- Executes on button press in NextImage.
function NextImage_Callback(hObject, eventdata, handles)
% hObject handle to NextImage (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
global Nimg sfiles Imgnow
nf = length(sfiles);
if nf>0
Nimg = Nimg+1;
if Nimg>length(sfiles)
Nimg = 1;
end
Img = imread(sfiles(Nimg).name);
Imgnow = Img;
subplot(1,1,1); imshow(Img)
title(sfiles(Nimg).name)
end
% --- Executes on button press in MultipleSegmentation.
function MultipleSegmentation_Callback(hObject, eventdata, handles)
% hObject handle to MultipleSegmentation (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of MultipleSegmentation
global mseg
mseg = 1-mseg;
% --- Executes on button press in MinusSegmentation.
function MinusSegmentation_Callback(hObject, eventdata, handles)
% hObject handle to MinusSegmentation (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
global pseg mseg seg Imgnow
pseg = pseg+0.05;
PlotImage(Imgnow,seg,pseg,mseg)
% --- Executes on button press in PlusSegmentation.
function PlusSegmentation_Callback(hObject, eventdata, handles)
% hObject handle to PlusSegmentation (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
global pseg mseg seg Imgnow
pseg = pseg-0.05;
PlotImage(Imgnow,seg,pseg,mseg)
% --- Executes on button press in DoSegmentation.
function DoSegmentation_Callback(hObject, eventdata, handles)
% hObject handle to DoSegmentation (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
global Imgnow pseg seg mseg
PlotImage(Imgnow,seg,pseg,mseg);
% --- Executes on button press in IntHOG.
function IntHOG_Callback(hObject, eventdata, handles)
% hObject handle to IntHOG (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Hint: get(hObject,'Value') returns toggle state of IntHOG
global fextractor
fextractor.inthog = get(hObject,'Value');
|
github
|
domingomery/Balu-master
|
Bfx_fourier.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_fourier.m
| 1,916 |
utf_8
|
a747d156cb43fd117284995b29b025b5
|
% [X,Xn,Xu] = Xfourier(I,R,options)
% [X,Xn,Xu] = Xfourier(I,options)
%
% Toolbox Xvis: Fourier features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% Example:
% options.Nfourier = 64; % imresize vertical
% options.Mfourier = 64; % imresize horizontal
% options.mfourier = 2; % imresize frequency vertical
% options.nfourier = 2; % imresize frequency horizontal
% options.show = 1; % display results
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = double(I(:,:,2))/256; % normalized green channel
% [X,Xn] = Xfourier(J,R,options); % Fourier features
% Bio_printfeatures(X,Xn)
%
% See also Xharalick, Xclp, Xgabor, Xdct, Xlbp.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Xfourier(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
I(R==0) = 0;
N = options.Nfourier;
M = options.Mfourier;
n = options.nfourier;
m = options.mfourier;
N2 = round(N/2);
M2 = round(M/2);
if options.show
disp('--- extracting Fourier features...');
end
Im = imresize(double(I),[N M]);
FIm = fft2(Im);
x = abs(FIm);
F = imresize(x(1:N2,1:M2),[n m]);
x = angle(FIm);
A = imresize(x(1:N2,1:M2),[n m]);
LS = 2*n*m;
X = zeros(1,LS);
Xn = char(zeros(LS,24));
k = 0;
for i=1:n
for j=1:m
k = k + 1;
s = sprintf('Fourier Abs (%d,%d) ',i,j);
Xn(k,:) = s(1:24);
X(k) = F(i,j);
end
end
for i=1:n
for j=1:m
k = k + 1;
s = sprintf('Fourier Ang (%d,%d)[rad] ',i,j);
Xn(k,:) = s(1:24);
X(k) = A(i,j);
end
end
|
github
|
domingomery/Balu-master
|
Bfx_hogi.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_hogi.m
| 3,334 |
utf_8
|
e85623c068f0196bd52a35b3453bb979
|
% [X,Xn] = Bfx_hog(I,options)
%
% Toolbox: Balu
% Histogram of Orientated Gradients features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% options.nj; : number of HOG windows per bound box
% options.ni : in i (vertical) and j (horizaontal) direction
% options.B : number of histogram bins
% options.show : show histograms (glyphs)
%
% Example:
% options.nj = 20; % 10 x 20
% options.ni = 10; % histograms
% options.B = 9; % 9 bins
% options.show = 1; % number of neighbor samples
% I = imread('testimg1.jpg'); % input image
% J = rgb2gray(I);
% figure(1);imshow(J,[]);
% figure(2);
% [X,Xn] = Bfx_hog(J,options); % HOG features (see gradients
% % arround perimeter.
%
% See also Bfx_phog, Bfx_lbp.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_hogi(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'normalize')
options.normalize = 0;
end
nj = options.nj; % number of HOG windows per bound box
ni = options.ni; % in i (vertical) and j (horizaontal) direction
B = options.B; % number of histogram bins
show = options.show; % show histograms
N = size(I,1);
M = size(I,2);
X = zeros(1,nj*ni*B); % column vector with zeros
I = double(I);
dj = floor(M/(nj+1));
di = floor(N/(ni+1));
t = 0;
%hj = [-1,0,1];
%hi = -hj';
%Gj = imfilter(I,hj);
%Gi = imfilter(I,hi);
%A = atan2(Gi,Gj);
%A = mod(A,pi);
%G = ((Gi.^2)+(Gj.^2)).^.5;
G = I(:,:,1)/255*363;
A = I(:,:,2)/255*pi;
K = 4*di*dj;
ang = zeros(K,1);
mag = zeros(K,1);
if show
J = zeros(N,M);
ss = min([N/ni M/nj])*0.40;
end
w = zeros(ni,nj,B);
for i = 1:ni
ii = (i-1)*di+1:(i+1)*di;
i0 = mean(ii);
for j = 1:nj
jj = (j-1)*dj+1:(j+1)*dj;
j0 = mean(jj);
t = t+1;
ang(:) = A(ii,jj);
mag(:) = G(ii,jj);
X2 = zeros(B,1);
for b = 1:B
q = find(ang<=pi*b/B);
X2(b) = X2(b)+sum(mag(q));
ang(q) = 5;
end
X2 = X2/(norm(X2)+0.01);
X(1,indices(t,B)) = X2;
% w(i,j,:) = X2;
if show
for b=1:B
alpha = pi*b/B;
q = -ss:ss;
qi = round(i0+q*cos(alpha));
qj = round(j0+q*sin(alpha));
qk = qi+(qj-1)*N;
J(qk) = J(qk)+X2(b);
end
J(round(i0),round(j0))=1;
end
end
end
if show
imshow(round(J/max2(J)*256),jet)
options.J = J;
end
options.w = w;
Xn = char(zeros(nj*ni*B,24));
Xn(:,1) = 'H'; Xn(:,2) = 'O'; Xn(:,3) = 'G';
J = uint8(zeros(N,M,3));
G(G>363) = 363; % max2(Gi) = max2(Gi) = 256 => max2(G) = sqrt(2*256^2)
J(:,:,1) = uint8(round(G/363*255));
J(:,:,2) = uint8(round(A/pi*255));
options.Ihog = J;
if options.normalize
X = X/sum(X);
end
|
github
|
domingomery/Balu-master
|
Bfx_fourierdes.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_fourierdes.m
| 1,997 |
utf_8
|
3026250f26d693f4384307cb01b351cf
|
% function [X,Xn] = Bfx_fourierdes(R,options)
%
% Toolbox: Balu
% Computes the Fourier descriptors of a binary image R.
%
% options.show = 1 display mesagges.
% options.Nfourierdes number of descriptors.
%
% X is the feature vector
% Xn is the list of feature names.
%
% Reference:
% Zahn, C; Roskies, R.: Fourier Descriptors for Plane
% Closed Curves, IEEE Trans on Computers, C21(3):269-281, 1972
%
% Example:
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_fourierdes(R); % Fourier descriptors
% Bio_printfeatures(X,Xn)
%
% See also Bfx_fitellipse, Bfx_hugeo, Bfx_gupta, Bfx_flusser.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_fourierdes(R,options)
if ~exist('options','var')
options.show = 0;
options.Nfourierdes = 16;
end
if options.show == 1
disp('--- extracting Fourier descriptors...');
end
N = options.Nfourierdes;
jj = sqrt(-1);
B = bwboundaries(R,'noholes');
g = B{1};
V = g(:,2)+jj*g(:,1);
m = size(g,1);
r = zeros(m,1);
phi = zeros(m,1);
dphi = zeros(m,1);
l = zeros(m,1);
dl = zeros(m,1);
r(1) = V(1)-V(m);
for i=2:m
r(i) = V(i)-V(i-1);
end
for i=1:m
dl(i) = abs(r(i));
phi(i) = angle(r(i));
end
for i=1:m-1
dphi(i) = mod(phi(i+1)-phi(i)+pi,2*pi)-pi;
end
dphi(m) = mod(phi(1)-phi(m)+pi,2*pi)-pi;
for k=1:m
l(k) = 0;
for i=1:k
l(k) = l(k) + dl(i);
end
end
L = l(m);
A = zeros(N,1);
for n=1:N
an = 0;
bn = 0;
for k = 1:m
an = an + dphi(k)*sin(2*pi*n*l(k)/L);
bn = bn + dphi(k)*cos(2*pi*n*l(k)/L);
end
an = -an/n/pi;
bn = bn/n/pi;
imagi = an + jj*bn;
A(n) = abs(imagi);
end
X = A';
Xn = char(zeros(N,24));
for i = 1:N
Xn(i,:) = sprintf('Fourier-des %2d ',i);
end
|
github
|
domingomery/Balu-master
|
Bfx_contrast.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_contrast.m
| 4,036 |
utf_8
|
8019b18ebe9a330ead03c4f53d925a0c
|
% [X,Xn] = Bfi_contrast(I,R,options)
% [X,Xn] = Bfi_contrast(I,options)
%
% Toolbox: Balu
% Contrast features.
%
% X is the features vector, Xn is the list of feature names(see Example
% to see how it works).
%
% References:
% Mery, D.; Filbert: Classification of Potential Defects in
% Automated Inspection of Aluminium Castings Using Statistical Pattern
% Recognition. In Proceedings of 8th European Conference on Non-
% Destructive Testing (ECNDT 2002), Jun. 17-21, 2002, Barcelona, Spain.
%
% Kamm, K.-F. Ewen, K. (ed.): Grundlagen der R?ntgenabbildung Moderne
% Bildgebung: Physik, Ger?tetechnik, Bildbearbeitung und -kommunikation,
% Strahlenschutz, Qualit?tskontrolle, Georg Thieme Verlag, 1998, 45-62
%
% Example:
% options.show = 1; % display results
% options.neighbor = 2; % neigborhood is imdilate
% options.param = 5; % with 5x5 mask
% I = imread('testimg4.jpg'); % input image
% J = I(395:425,415:442,1); % region of interest (red)
% R = J<=130; % segmentation
% figure;imshow(J,[])
% figure;imshow(R)
% [X,Xn] = Bfx_contrast(J,R,options); % contrast features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_clp, Bfx_haralick, Bfx_contrast, Bfx_fourier, Bfx_dct, Bfx_lbp.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_contrast(I,R,options)
I = double(I);
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'show')
options.show = 0;
end
if options.show == 1
disp('--- extracting contrast features...');
end
s = options.param;
switch options.neighbor
case 1
[N,M] = size(R);
[ii,jj] = find(R==1);
i_m = mean(ii);
j_m = mean(jj);
id = (double(max(ii)-min(ii)+1)*s/2);
jd = (double(max(jj)-min(jj)+1)*s/2);
i1 = max([1 round(i_m-id)]);
i2 = min([N round(i_m+id)]);
j1 = max([1 round(j_m-jd)]);
j2 = min([M round(j_m+jd)]);
Rn = zeros(size(R));
Rn(i1:i2,j1:j2)=1;
case 2
Rn = imdilate(R,ones(s,s));
[ii,jj] = find(Rn==1);
i1 = min(ii);
i2 = max(ii);
j1 = min(jj);
j2 = max(jj);
end
Rn = and(Rn,not(R));
if sum(Rn(:)) > 0
MeanGr = mean(I(R==1));
MeanGn = mean(I(Rn==1));
K1 = (MeanGr-MeanGn)/MeanGn; % contrast after Kamm, 1999
K2 = (MeanGr-MeanGn)/(MeanGr+MeanGn); % modulation after Kamm, 1999
K3 = log(MeanGr/MeanGn); % film-contrast after Kamm, 1999
else
K1 = -1;
K2 = -1;
K3 = -1;
end
[Ks,K] = contrast2002(I(i1:i2,j1:j2)); % contrast after Mery Barcelona 2002 XRA0152
X = [K1 K2 K3 Ks K];
Xn = [ 'contrast-K1 '
'contrast-K2 '
'contrast-K3 '
'contrast-Ks '
'contrast-K '];
end
% [Ks,K] = contrast2002(I)
%
% Toolbox: Balu
% Contrast features after Mery & Filbert, 2002.
%
% D.Mery, PUC-DCC, Apr. 2008
% http://dmery.ing.puc.cl
function [Ks,K] = contrast2002(I)
[nI,mI] = size(I);
n1 = fix(nI/2)+1;
m1 = fix(mI/2)+1;
P1 = I(n1,:); % Profile in i-Direction
P2 = I(:,m1)'; % Profile in j-Direction
Q1 = rampefr(P1); % Profile P1 without ramp
Q2 = rampefr(P2); % Profile P2 without ramp
Q = [Q1 Q2]; % Fusion of profiles
Ks = std(Q); % Contrast Ks
K = log(max(Q)-min(Q)+1); % Contrast K
end
% Q = rampefr(P)
%
% Toolbox: Balu
% Eliminate ramp of profile P (used to compute contrast features).
%
% D.Mery, PUC-DCC, Apr. 2008
% http://dmery.ing.puc.cl
function Q = rampefr(P)
k = length(P);
m = (P(k)-P(1))/(k-1);
b = P(1)-m;
Q = P - (1:k)*m - b*ones(1,k);
end
|
github
|
domingomery/Balu-master
|
Bfx_bsif.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_bsif.m
| 4,662 |
utf_8
|
ff6279e64bd0b6798ec5b8c1e69d967f
|
% [X,Xn,options] = Bfx_bsif(I,R,options)
% [X,Xn,options] = Bfx_bsif(I,options)
% [X,Xn] = Bfx_bsif(I,R,options)
% [X,Xn] = Bfx_bsif(I,options)
%
% Toolbox: Balu
% Binarized statistical image features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% It calculates the BSIF over the a regular grid of patches. The function
% uses Juho Kannala and Esa Rahtu (see http://www.ee.oulu.fi/~jkannala/bsif/bsif.html).
%
% It returns a matrix of uniform bsif descriptors for I, made by
% concatenating histograms of each grid cell in the image.
% Grid size is options.hdiv * options.vdiv
%
% R is a binary image or empty. If R is given the bsif will be computed
% the corresponding pixles R==0 in image I will be set to 0.
%
% Output: %%%%%%%%%%%%%%%%%%%%%revisar%%%%%%%%%%%%
% X is a matrix of size ((hdiv*vdiv) x 59), each row has a
% histogram corresponding to a grid cell. We use 59 bins.
% options.x of size hdiv*vdiv is the x coordinates of center of ith grid cell
% options.y of size hdiv*vdiv is the y coordinates of center of ith grid cell
% Both coordinates are calculated as if image was a square of side length 1.
%
% References:
% J. Kannala and E. Rahtu. Bsif: Binarized statistical image features.
% In Pattern Recognition (ICPR), 2012 21st International Conference on,
% pages 1363--1366. IEEE, 2012.
%
%
% Example 1:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.filter = 7; % use filter 7x7
% options.bits = 11; % use 11 bits filter
% options.mode = 'h'; % return histogram
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_bsif(J,[],options); % BSIF features
% figure(2);bar(X) % histogram
%
% Example 2:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.filter = 7; % use filter 7x7
% options.bits = 11; % use 11 bits filter
% options.mode = 'nh'; % return normilized histogram
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_bsif(J,[],options); % BSIF features
% figure(2);bar(X) % histogram
%
% Example 3:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.filter = 7; % use filter 7x7
% options.bits = 11; % use 11 bits filter
% options.mode = 'im'; % return image represetation
% %(image is only aviable for vdiv = 1 y hdiv = 1)
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_bsif(J,[],options); % BSIF features
% figure(2);imshow(X,[]); % display image
%
%
% See also Bfx_lbp, Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct.
%
% (c) Erick Svec
%
function [X,Xn] = Bfx_bsif(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
vdiv = options.vdiv;
hdiv = options.hdiv;
if ~isfield(options,'show')
options.show = 0;
end
if options.show == 1
disp('--- extracting binarized statistical image features...');
end
filename=['ICAtextureFilters_' int2str(options.filter) 'x' int2str(options.filter) '_' int2str(options.bits) 'bit'];
load(filename, 'ICAtextureFilters');
if ~isempty(R);
I(R==0) = 0;
end
if strcmp(options.mode,'im')
if vdiv == 1 && hdiv == 1
X = bsif(I,ICAtextureFilters,options.mode);
Xn = options;
return
else
throw(MException('MATLAB:odearguments:InconsistentDataType', 'Invalid Options Set: vdiv and hdiv must be 1 for mode im (when try retrieve image representation)'));
return
end
end
[N,M] = size(I);
w = N/vdiv;
h = M/hdiv;
X = [];
Xn = options;
for i = 1:vdiv
for j = 1:hdiv
part = I(i*w-w+1:w*i,j*h-h+1:h*j);
code_img = bsif(part,ICAtextureFilters,options.mode);
X = [X code_img];
end
end
end
|
github
|
domingomery/Balu-master
|
Bfx_randomsliwin.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_randomsliwin.m
| 5,864 |
utf_8
|
e8ce09e5cd7d838b6159ae1aa8b838ff
|
% [X,d,Xn,x] = Bfx_randomsliwin(I,J,options)
%
% Toolbox: Balu
%
% Feature extraction of random sliding windows.
% This program select automatically detection windows sized mxm
% with label '1' and lable '0'. For each window
% Balu intensity features are extracted.
%
% Input:
% I original image (more than one channel is allowed)
% J ideal segmentation
% options.opf feature extraction options (see example)
% options.selec selected features, selec = 0 means all features
% options.m sliding window size in pixels (mxm)
% options.n0 number of '0' windows
% options.n1 number of '1' windows
% options.ROI region of interest where the windows are extracted
% options.th0 if the number of '1' in detection window/m^2 < th0 a '0'
% sample is selected
% options.th1 if the number of '1' in detection window/m^2 >=th1 a '1'
% sample is selected
% options.show display detected windows
%
% Output:
% X feature values
% Xn feature names
% d ideal classification (0 or 1) of each sample
% x ceter of mass (j,i) of each patch
%
% Example:
% I1 = imread('testimg7.bmp'); % grayvalue image
% [I_on,I2] = Bim_cssalient(I1,1,0); % saliency map
% I(:,:,1) = I1; % channel 1
% I(:,:,2) = I2; % cahnnel 2
% J = imread('testimg8.bmp'); % ideal segmentation
% bf(1).name = 'lbp'; % definition of
% bf(1).options.show = 0; % first features
% bf(1).options.vdiv = 1;
% bf(1).options.hdiv = 1;
% bf(2).name = 'basicint'; % definition of
% bf(2).options.show = 0; % second features
% bf(2).options.mask = 5;
% opf.b = bf;
% opf.colstr = 'gs'; % chn 1,2 are gray,sal
% options.opf = opf;
% options.selec = 0; % all features
% options.m = 24; % size of a window mxm
% options.n0 = 100; % number of 0 windows
% options.n1 = 100; % number of 1 windows
% options.th0 = 0.02; % threshold for 0
% options.th1 = 0.02; % threshold for 1
% options.show = 1;
% [X,d,Xn] = Bfx_randomsliwin(I,J,options);
%
% See also Bim_segsliwin.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,d,Xn,x] = Bfx_randomsliwin(I,J,options)
warning off
opf = options.opf;
% selec = options.selec;
m = options.m; % size of a window mxm
n0 = options.n0;
n1 = options.n1;
th0 = options.th0;
th1 = options.th1;
show = options.show;
if isfield(options,'win');
win = options.win;
else
win = 30;
end
if isfield(options,'roi');
ROI = options.roi;
else
ROI = ones(size(I));
end
if isfield(options,'selec');
selec = options.selec;
else
selec = 0;
end
[N,M]=size(I(:,:,1));
if show==1
close all
figure(1)
imshow(I(:,:,1),[]);title('Original');
pause(10)
hold on
%figure(2)
%imshow(J);title('Real Defects');
%figure(1)
end
m1 = m-1;
md = (m1-1)/2;
m2 = m*m;
R = ones(m,m);
if show~=-1
ff = Bio_statusbar('Extracting patches');
end
ft = Bfx_int(I(1:win,1:win),opf);
nf = size(ft,2);
% f = [];
% x = [];
nn = n0+n1;
f = zeros(nn,nf);
x = zeros(nn,2);
d = [zeros(n0,1); ones(n1,1)];
k = 0;
if n0>0
% Extracting features for '0' detection windows
if show==1
disp('Extracting features for 0 detection windows...');
end
i = 0;
while i<n0
i1 = fix(rand*(N-m1))+1;
j1 = fix(rand*(M-m1))+1;
rj = sum2(ROI(i1:i1+m1,j1:j1+m1))/m2;
if rj>0.90
wj = J(i1:i1+m1,j1:j1+m1);
t = sum(wj(:))/m2;
if t<th0
wij = I(i1:i1+m1,j1:j1+m1,:);
[ft,fn] = Bfx_int(wij,R,opf);
% f = [f;ft 0];
% x = [x;j1+md i1+md];
k = k+1;
f(k,:) = ft;
x(k,:) = [j1+md i1+md];
i = i+1;
if show~=-1
ff = Bio_statusbar(k/nn,ff);
end
if show==1
% fprintf('0: %d/%d\n',i,n0);
plot([j1 j1 j1+m1 j1+m1 j1],[i1 i1+m1 i1+m1 i1 i1],'g')
drawnow
end
end
end
end
end
if n1>0
% Extracting features for '1' detection windows
if show==1
disp('Extracting features for 1 detection windows...');
end
i = 0;
while i<n1
i1 = fix(rand*(N-m1))+1;
j1 = fix(rand*(M-m1))+1;
rj = sum2(ROI(i1:i1+m1,j1:j1+m1))/m2;
if rj>0.95
wj = J(i1:i1+m1,j1:j1+m1);
t = sum(wj(:))/m2;
if t>=th1
wij = I(i1:i1+m1,j1:j1+m1,:);
[ft,fn] = Bfx_int(wij,R,opf);
% f = [f;ft 1];
% x = [x;j1+md i1+md];
k = k+1;
f(k,:) = ft;
x(k,:) = [j1+md i1+md];
i = i+1;
if show~=-1
ff = Bio_statusbar(k/nn,ff);
end
if show==1
% fprintf('1: %d/%d\n',i,n1);
plot([j1 j1 j1+m1 j1+m1 j1],[i1 i1+m1 i1+m1 i1 i1],'r')
drawnow
end
end
end
end
end
if sum(selec)==0
X = f;
Xn = fn;
else
X = f(:,selec);
Xn = fn(selec,:);
end
if show~=-1
delete(ff);
end
|
github
|
domingomery/Balu-master
|
Bfx_lbphog.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbphog.m
| 1,498 |
utf_8
|
35bc94d93b874692e58084e3c10a81e7
|
% [X,Xn] = Bfx_hog(I,options)
%
% Toolbox: Balu
% Histogram of Orientated Gradients features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% options.nj; : number of HOG windows per bound box
% options.ni : in i (vertical) and j (horizaontal) direction
% options.B : number of histogram bins
% options.show : show histograms (glyphs)
%
% Example:
% options.nj = 20; % 10 x 20
% options.ni = 10; % histograms
% options.B = 9; % 9 bins
% options.show = 1; % number of neighbor samples
% I = imread('testimg1.jpg'); % input image
% J = rgb2gray(I);
% figure(1);imshow(J,[]);
% figure(2);
% [X,Xn] = Bfx_hog(J,options); % HOG features (see gradients
% % arround perimeter.
%
% See also Bfx_phog, Bfx_lbp.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_lbphog(I,R,options)
if nargin==2;
options = R;
R = [];
end
[X_lbp,Xn_lbp,op_lbp] = Bfx_lbp(I,R,options);
[X_hog,Xn_hog,op_hog] = Bfx_hog(I,R,options);
I_lbp = op_lbp.Ilbp;
I_hog = op_hog.Ihog;
X = [X_lbp X_hog ];
Xn = [Xn_lbp; Xn_hog];
N = size(I,1);
M = size(I,2);
J = uint8(zeros(N,M,3));
J(:,:,1) = I_lbp;
J(:,:,2) = I_hog(:,:,1);
J(:,:,3) = I_hog(:,:,2);
options.Ilbphog = J;
|
github
|
domingomery/Balu-master
|
Bfx_vlhog.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_vlhog.m
| 1,557 |
utf_8
|
d708218a0d2591812a418a1ac0eadbdc
|
% [X,Xn] = Bfx_vlhog(I,options)
%
% Toolbox: Balu
% Histogram of Orientated Gradients features using Vlfeat Toolbox.
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% options.cellsize : size of the cells in pixels
% options.variant : 1 for UoCTTI, 2 for Dalal-Triggs
% options.show : 1 shows oriented histograms
%
% Example:
% options.cellsize = 32; % 32 x 32
% options.variant = 1; % UoCTTI
% options.show = 1; % show results
% I = imread('testimg1.jpg'); % input image
% J = rgb2gray(I);
% figure(1);imshow(J,[]);
% figure(2);
% [X,Xn] = Bfx_vlhog(J,options); % HOG features (see gradients
% % arround perimeter).
%
% See also Bfx_phog, Bfx_lbp, Bfx_hog, vl_hog.
%
% (c) GRIMA-DCCUC, 2012
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_vlhog(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'variant')
options.varian = 1;
end
if ~isfield(options,'show')
options.show = 1;
end
if options.variant == 1
varname = 'UoCTTI';
else
varname = 'DalalTriggs';
end
options.hog = vl_hog(im2single(I),options.cellsize,'variant',varname);
if options.show==1
figure
options.Ir = vl_hog('render',options.hog);
imshow(options.Ir,[]);
end
X = options.hog(:)';
n = length(X);
Xn = zeros(n,24);
|
github
|
domingomery/Balu-master
|
Bfx_fitellipse.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_fitellipse.m
| 3,128 |
utf_8
|
a1b463a5bd9a20f8037bd3f60f769d2c
|
% [X,Xn] = Bfx_fitellipse(R,options)
% [X,Xn] = Bfx_fitellipse(R)
%
% Toolbox: Balu
%
% Fit ellipse for the boundary of a binary image R.
%
% options.show = 1 display mesagges.
%
% X is a 6 elements vector:
% X(1): Ellipse-centre i direction
% X(2): Ellipse-centre j direction
% X(3): Ellipse-minor axis
% X(4): Ellipse-major axis
% X(5): Ellipse-orientation
% X(6): Ellipse-eccentricity
% X(7): Ellipse-area
%
% Xn is the list of feature names.
%
% Xn is the list of feature names.
%
% Reference:
% Fitzgibbon, A.; Pilu, M. & Fisher, R.B. (1999): Direct Least Square
% Fitting Ellipses, IEEE Trans. Pattern Analysis and Machine
% Intelligence, 21(5): 476-480.
%
% Example:
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_fitellipse(R); % ellipse features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_basicgeo, Bfx_hugeo, Bfx_gupta, Bfx_flusser.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_fitellipse(R,options)
E = bwperim(R,4);
[Y,X] = find(E==1); % pixel of perimeter in (i,j)
if length(X)>5
if ~exist('options','var')
options.show = 0;
end
if options.show == 1
disp('--- extracting ellipse features...');
end
% normalize data
mx = mean(X);
my = mean(Y);
sx = (max(X)-min(X))/2;
sy = (max(Y)-min(Y))/2;
x = (X-mx)/sx;
y = (Y-my)/sy;
% Build design matrix
D = [ x.*x x.*y y.*y x y ones(size(x)) ];
[U,S,V] = svd(D);
A = V(:,6);
% unnormalize
a = [
A(1)*sy*sy, ...
A(2)*sx*sy, ...
A(3)*sx*sx, ...
-2*A(1)*sy*sy*mx - A(2)*sx*sy*my + A(4)*sx*sy*sy, ...
-A(2)*sx*sy*mx - 2*A(3)*sx*sx*my + A(5)*sx*sx*sy, ...
A(1)*sy*sy*mx*mx + A(2)*sx*sy*mx*my + A(3)*sx*sx*my*my ...
- A(4)*sx*sy*sy*mx - A(5)*sx*sx*sy*my ...
+ A(6)*sx*sx*sy*sy ...
]';
a = a/a(6);
% get ellipse orientation
alpha = atan2(a(2),a(1)-a(3))/2;
% get scaled major/minor axes
ct = cos(alpha);
st = sin(alpha);
ap = a(1)*ct*ct + a(2)*ct*st + a(3)*st*st;
cp = a(1)*st*st - a(2)*ct*st + a(3)*ct*ct;
% get translations
T = [[a(1) a(2)/2]' [a(2)/2 a(3)]'];
mc = -inv(2*T)*[a(4) a(5)]';
% get scale factor
val = mc'*T*mc;
scale = abs(1 / (val- a(6)));
% get major/minor axis radii
ae = 1/sqrt(scale*abs(ap));
be = 1/sqrt(scale*abs(cp));
ecc = ae/be; % eccentricity
ar = pi*ae*be;
X = [ mc' ae be alpha ecc ar];
else
X = [0 0 0 0 0 0 0];
disp('Warning: Bfx_fitellipse does not have enough points to fit');
end
Xn = [
'Ellipse-centre i [px] '
'Ellipse-centre j [px] '
'Ellipse-minor ax [px] '
'Ellipse-major ax [px] '
'Ellipse-orient [rad] '
'Ellipse-eccentricity '
'Ellipse-area [px] '
];
|
github
|
domingomery/Balu-master
|
Bfx_files.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_files.m
| 8,658 |
utf_8
|
b57e06bded6b2d40a2bc7376dab47b76
|
% [X,Xn,S] = Bfx_files(f,opf) % gemetric and intensity features
% [X,Xn,S] = Bfx_files(f,opf,labelling) % features + labelling
%
% Toolbox: Balu
%
% Feature extraction from a set of files.
%
% This function calls feature extraction procedures of all
% images defined in f. See example to see how it works.
%
% X is the feature matrix (one feature per column, one sample per row),
% Xn is the list of feature names (see Example to see how it works).
%
% S is the list of filenames of the images. The features of file S(i,:)
% are in row X(i,:).
%
% Example:
% f.path = ''; % current directory or a path directory
% f.prefix = 'testimg'; f.extension = '.jpg';
% f.digits = 1;
% f.gray = 0;
% f.subsample = 1;
% f.resize = 1;
% f.imgmin = 1;
% f.imgmax = 2;
% f.window = [];
% f.negative = 0;
% f.sequence = 1:f.imgmax;
%
% b(1).name = 'gabor'; b(1).options.show=1; % Gabor features
% b(1).options.Lgabor = 8; % number of rotations
% b(1).options.Sgabor = 8; % number of dilations (scale)
% b(1).options.fhgabor = 2; % highest frequency of interest
% b(1).options.flgabor = 0.1; % lowest frequency of interest
% b(1).options.Mgabor = 21; % mask size
% b(1).options.type = 2; % intensity
%
% b(2).name = 'basicint'; b(2).options.show = 1; % Basic intensity features
% b(2).options.type = 2; % intensity
%
% b(3).name = 'lbp'; b(3).options.show = 1; % Fourier
% b(3).options.vdiv = 2; % vertical div
% b(3).options.hdiv = 2; % horizontal div
% b(3).options.type = 2; % intensity
%
%
% b(4).name = 'hugeo'; b(4).options.show = 1; % Hu moments
% b(4).options.type = 1; % geometric
%
% b(5).name = 'flusser'; b(5).options.show = 1; % Flusser moments
% b(5).options.type = 1; % geometric
%
% b(6).name = 'fourierdes'; b(6).options.show = 1; % Fourier
% b(6).options.Nfourierdes=12; % descriptors
% b(6).options.type = 1; % geometric
%
% opf.b = b;
% opf.channels = 'RGB'; % RGB images
% opf.segmentation = 'Bim_segbalu'; % segmentation
% opf.param = -0.05; % parameters of segmentation
% opf.intensity = 1;
%
% [X,Xn,S] = Bfx_files(f,opf);
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,Xn,S,d] = Bfx_files(f,opf,labeling)
if f.imgmax == 0
error('Bfx_files: set of images is empty.')
end
if ~exist('labeling','var')
labeling = 0;
end
d = zeros(f.imgmax-f.imgmin+1,1);
if compare(class(opf.b),'cell')==0
opf.b = Bfx_build(opf.b);
end
if isfield(opf,'segmentation')
doseg = 1;
seg = opf.segmentation;
if opf.segmentation == 0
doseg = 0;
end
else
doseg = 0;
end
n = length(opf.b);
kg = 0;
ki = 0;
opg = [];
opi = [];
for i=1:n
if opf.b(i).options.type == 1
kg = kg+1;
opg.b(kg) = opf.b(i);
else
ki = ki+1;
opi.b(ki) = opf.b(i);
end
end
X = [];
S = [];
if doseg
if isfield(opf,'param')
par = opf.param;
else
par = [];
end
end
if isfield(opf,'intensity')
inten = opf.intensity;
else
inten = doseg;
end
ct = 'gRGBHSVLab12';
co = zeros(length(ct),1);
if isfield(opf,'channels')
ch = opf.channels;
if compare(class(ch),'char')==0
colorstr = ch;
else % cell
colorstr = [];
nx = length(ch);
for i=1:nx
chs = lower(char(ch(i)));
switch chs
case 'gray'
str = 'g';
case 'red'
str = 'R';
case 'green'
str = 'G';
case 'blue'
str = 'B';
case 'hue'
str = 'H';
case {'sat','saturation'}
str = 'S';
case {'value','val'}
str = 'V';
case {'l','l*'}
str = 'L';
case {'a','a*'}
str = 'a';
case {'b','b*'}
str = 'b';
case {'saliency_1','saliency_on'}
str = '1';
case {'saliency_2','saliency_off'}
str = '2';
end
colorstr = [colorstr str];
end
end
else
colorstr = 'g';
end
nc = length(colorstr);
for i=1:nc
opi.colstr = colorstr;
ii = ct==colorstr(i);
co(ii) = 1;
end
nc = sum(co);
if nc>0
ff = Bio_statusbar('Feature Extraction');
opf.channels = ct(co==1);
for i=f.imgmin:f.imgmax
ff = Bio_statusbar((i-f.imgmin)/(f.imgmax-f.imgmin+1),ff);
Xi = [];
Xn = [];
[I,st] = Bio_loadimg(f,i);
if labeling
imshow(I(:,:,1),[])
end
[N,M,P] = size(I);
IX = zeros(N,M,nc);
if sum(co(2:4))>0
XRGB = I;
end
if sum(co(5:7))>0
XHSV = rgb2hsv(I);
end
if sum(co(8:10))>0
if (exist('LAB.mat','file'))
load LAB
XLAB = Bim_rgb2lab(I,M);
else
disp('Warning: LAB.mat does not exist. CIE formulas will be used for L*a*b conversion');
XLAB = Bim_rgb2lab0(I);
end
end
if sum(co(11:12))>0
[J_on,J_off] = Bim_cssalient(I,1,0);
end
k = 0;
jj = find(co);
for ji=1:length(jj)
j = jj(ji);
k = k + 1;
switch j
case 1 % Gray
if size(I,3)==3
IX(:,:,k) = rgb2gray(I/256)*256;
else
if max(I(:))>256
IX(:,:,k) = I/256;
else
IX(:,:,k) = I;
end
end
case 2 % Red
IX(:,:,k) = XRGB(:,:,1);
case 3 % Green
IX(:,:,k) = XRGB(:,:,2);
case 4 % Blue
IX(:,:,k) = XRGB(:,:,3);
case 5 % H
IX(:,:,k) = XHSV(:,:,1);
case 6 % S
IX(:,:,k) = XHSV(:,:,2);
case 7 % V
IX(:,:,k) = XHSV(:,:,3);
case 8 % L*
IX(:,:,k) = XLAB(:,:,1);
case 9 % a*
IX(:,:,k) = XLAB(:,:,2);
case 10 % b*
IX(:,:,k) = XLAB(:,:,3);
case 11 % Saliency on
IX(:,:,k) = J_on;
case 12 % Saliency on
IX(:,:,k) = J_off;
end
%end
end
fprintf('\n--- processing image %s...\n',st);
if doseg
if ~isempty(par)
Rg = feval(seg,I,par);
else
Rg = feval(seg,I);
end
else
Rg = [];
end
if inten
Ri = Rg;
else
Ri = [];
end
if ~isempty(opg)
if ~isempty(opg.b)
[Xgeo,Xng] = Bfx_geo(Rg,opg);
Xi = [Xi Xgeo];
Xn = [Xn;Xng];
end
end
if ~isempty(opi)
if ~isempty(opi.b)
[Xint,Xni] = Bfx_int(IX,Ri,opi);
Xi = [Xi Xint];
Xn = [Xn;Xni];
end
end
if labeling
d(i-f.imgmin+1,1) = input('Label for this image? ');
end
X = [X;Xi];
st = [st ones(1,200)*' '];
S = [S;st(1:100)];
end
delete(ff);
else
error('Bfx_files error: Colors %s are recognized',colorstr);
end
|
github
|
domingomery/Balu-master
|
Bfx_haralick.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_haralick.m
| 6,209 |
utf_8
|
08e5cceaeada8d46d9b1ced71e2004ce
|
% [X,Xn] = Bfx_haralick(I,R,options)
% [X,Xn] = Bfx_haralick(I,options)
%
% Toolbox: Balu
% Haralick texture features.
%
% X is a 28 elements vector with mean and range of mean and range of
%
% 1 Angular Second Moment
% 2 Contrast
% 3 Correlacion
% 4 Sum of squares
% 5 Inverse Difference Moment
% 6 Sum Average
% 8 Sum Entropy
% 7 Sum Variance
% 9 Entropy
% 10 Difference Variance
% 11 Difference Entropy
% 12,13 Information Measures of Correlation
% 14 Maximal Corrleation Coefficient
%
% Xn is the list of name features.
%
% I is the image. R is the binary image that indicates which pixels of I will be
% computed.
% options.dharalick is the distance in pixels used to compute the
% coocurrence matrix.
% options.show = 1 display results.
%
% Reference:
% Haralick (1979): Statistical and Structural Approaches to Texture,
% Proc. IEEE, 67(5):786-804
%
% Example 1: only one distance (3 pixels)
% options.dharalick = 3; % 3 pixels distance for coocurrence
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = I(:,:,2); % green channel
% [X,Xn] = Bfx_haralick(J,R,options); % Haralick features
% Bio_printfeatures(X,Xn)
%
% Example 2: five distances (1,2,...5 pixels)
% options.dharalick = 1:5; % 3 pixels distance for coocurrence
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = I(:,:,2); % green channel
% [X,Xn] = Bfx_haralick(J,R,options); % Haralick features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct, Bfx_lbp.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function [X,Xn] = Bfx_haralick(I,R,options)
I = double(I);
if nargin==2;
options = R;
R = ones(size(I));
end
if isempty(R)
R = ones(size(I));
end
dseq = options.dharalick;
if ~isfield(options,'show')
options.show = 0;
end
if options.show == 1
disp('--- extracting Haralick texture features...');
end
m = length(dseq);
n = 28*m;
X = zeros(1,n);
Xn = char(zeros(n,24));
k = 1;
for i=1:m
d = dseq(i);
Cd000 = Bcoocurrencematrix(I, R, d, 0)+Bcoocurrencematrix(I,R, -d, 0);Cd000 = Cd000/sum(Cd000(:));
Cd045 = Bcoocurrencematrix(I, R, d,-d)+Bcoocurrencematrix(I,R, -d, d);Cd045 = Cd045/sum(Cd045(:));
Cd090 = Bcoocurrencematrix(I, R, 0, d)+Bcoocurrencematrix(I,R, 0,-d);Cd090 = Cd090/sum(Cd090(:));
Cd135 = Bcoocurrencematrix(I, R, d, d)+Bcoocurrencematrix(I,R, -d,-d);Cd135 = Cd135/sum(Cd135(:));
TexMat = [Bcoocurrencefeatures(Cd000) Bcoocurrencefeatures(Cd045) Bcoocurrencefeatures(Cd090) Bcoocurrencefeatures(Cd135)];
X(1,i*28-27:i*28) = [mean(TexMat,2); max(abs(TexMat'))']';
for q=1:2
if (q==1)
sq = 'mean ';
else
sq = 'range';
end
for s=1:14
Xn(k,:) = sprintf('Tx%2d,d%2d(%s) ',s,d,sq);
k = k + 1;
end
end
end
end
% P = Bcoocurrencematrix(I,R,Io,Jo)
%
% Coocurrence matrix of the pixels of image I indicated by binary image R
% following the direction (Io,Jo).
%
% (c) D.Mery, PUC-DCC, Apr. 2008
function P = Bcoocurrencematrix(I,R,Io,Jo)
V = fix(I/32)+1;
[N,M] = size(I);
Z1 = zeros(N+40,M+40);
Z2 = Z1;
R1 = Z1;
R2 = R1;
Z1(15:N+14,15:M+14) = V;
Z2(15+Io:N+14+Io,15+Jo:M+14+Jo) = V;
R1(15:N+14,15:M+14) = R;
R2(15+Io:N+14+Io,15+Jo:M+14+Jo) = R;
ii = find(not(and(R1,R2)));
Z1(ii) = -ones(length(ii),1);
Z2(ii) = -ones(length(ii),1);
T1 = Z1(:);
T2 = Z2(:);
d = find(and((T1>-1),(T2>-1)));
if (not(isempty(d)))
P = zeros(8,8);
X = sortrows([T1(d) T2(d)]);
i1 = find(or(([0; X(:,1)]-[X(:,1); 0]~=0),...
([0; X(:,2)]-[X(:,2); 0]~=0)));
i2 = [i1(2:length(i1)); 0];
d = i2-i1;
for i=1:length(d)-1
P(X(i1(i),2),X(i1(i),1)) = d(i);
end
else
P = -ones(8,8);
end
end
% Tx = Bcoocurrencefeatures(P)
%
% Haralick texture features calculated from coocurrence matrix P.
%
% (c) D.Mery, PUC-DCC, Apr. 2008
function Tx = Bcoocurrencefeatures(P)
Pij = P(:);
Ng = 8;
pxi = sum(P,2);
pyj = sum(P)';
ux = mean(pxi);
uy = mean(pyj);
sx = std(pxi);
sy = std(pyj);
pxy1 = zeros(2*Ng-1,1);
for k=2:2*Ng
s = 0;
for i=1:Ng
for j=1:Ng
if (i+j == k)
s = s + P(i,j);
end
end
end
pxy1(k-1) = s;
end
pxy2 = zeros(Ng,1);
for k=0:Ng-1
s = 0;
for i=1:Ng
for j=1:Ng
if (abs(i-j) == k)
s = s + P(i,j);
end
end
end
pxy2(k+1) = s;
end
Q = zeros(Ng,Ng);
pxi = pxi+1e-20;
pyj = pyj+1e-20;
for i=1:Ng
for j=1:Ng
s = 0;
for k=1:Ng
s = s + P(i,k)*P(j,k)/pxi(i)/pyj(k);
end
Q(i,j) = s;
end
end
eigQ = eig(Q);
[i,j] = find(P>=0);
dif = i-j;
dif2 = dif.*dif;
dif21 = dif2 + 1;
% 1 Angular Second Moment
f1 = Pij'*Pij;
% 2 Contrast
f2 = ((0:Ng-1).*(0:Ng-1))*pxy2;
% 3 Correlacion
f3 = (sum(i.*j.*Pij)-ux*uy*Ng^2)/sx/sy;
% 4 Sum of squares
f4 = dif2'*Pij;
% 5 Inverse Difference Moment
f5 = sum(Pij./dif21);
% 6 Sum Average
f6 = (2:2*Ng)*pxy1;
% 8 Sum Entropy
f8 = -pxy1'*log(pxy1+1e-20);
% 7 Sum Variance
if8 = (2:2*Ng)'-f8;
f7 = if8'*pxy1;
% 9 Entropy
f9 = -Pij'*log(Pij+1e-20);
% 10 Difference Variance
f10 = var(pxy2);
% 11 Difference Entropy
f11 = -pxy2'*log(pxy2+1e-20);
% 12,13 Information Measures of Correlation
HXY = f9;
pxipyj = pxi(i).*pyj(j);
HXY1 = -Pij'*log(pxipyj+1e-20);
HXY2 = -pxipyj'*log(pxipyj+1e-20);
HX = -pxi'*log(pxi+1e-20);
HY = -pyj'*log(pyj+1e-20);
f12 = (HXY-HXY1)/max([HX HY]);
f13 = (1-exp(-2*(HXY2-HXY)));
% 14 Maximal Corrleation Coefficient
f14 = (eigQ(2));
Tx = [f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13 f14]';
end
|
github
|
domingomery/Balu-master
|
Bfx_dct.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_dct.m
| 1,807 |
utf_8
|
f7545cce1977db3f89d1e673978c198f
|
% [X,Xn,] = Bfx_dct(I,R,options)
% [X,Xn] = Bfx_dct(I,options)
%
% Toolbox: Balu
% DCT features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% Reference:
% Kumar, A.; Pang, G.K.H. (2002): Defect detection in textured materials
% using Gabor filters. IEEE Transactions on Industry Applications,
% 38(2):425-440.
%
% Example:
% options.Ndct = 64; % imresize vertical
% options.Mdct = 64; % imresize horizontal
% options.mdct = 2; % imresize frequency vertical
% options.ndct = 2; % imresize frequency horizontal
% options.show = 1; % display results
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = double(I(:,:,2))/256; % normalized green channel
% [X,Xn] = Bfx_dct(J,R,options); % dct features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_haralick, Bfx_clp, Bfx_gabor, Bfx_fourier, Bfx_lbp.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_dct(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
I(R==0) = 0;
N = options.Ndct;
M = options.Mdct;
n = options.ndct;
m = options.mdct;
N2 = round(N/2);
M2 = round(M/2);
if options.show
disp('--- extracting dct features...');
end
Im = imresize(double(I),[N M]);
Fm = abs(dct2(Im));
F = imresize(Fm(1:N2,1:M2),[n m]);
LS = n*m;
X = zeros(1,LS);
Xn = char(zeros(LS,24));
k = 0;
for i=1:n
for j=1:m
k = k + 1;
s = sprintf('DCT(%d,%d) ',i,j);
Xn(k,:) = s(1:24);
X(k) = F(i,j);
end
end
|
github
|
domingomery/Balu-master
|
Bfx_gaborfull.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_gaborfull.m
| 7,865 |
utf_8
|
175cda6c002b920179c6a48331bb5f13
|
% [X,Xn] = Bfx_gaborfull(I,R,options)
% [X,Xn] = Bfx_gaborfull(I,options)
%
% Toolbox: Balu
% Gabor Full features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% Reference:
% M. Haghighat, S. Zonouz, M. Abdel-Mottaleb, "CloudID: Trustworthy
% cloud-based and cross-enterprise biometric identification,"
% Expert Systems with Applications, vol. 42, no. 21, pp. 7905-7916, 2015.
%
% Example:
% options.d1 = 4; % factor of downsampling along rows.
% options.d2 = 4; % factor of downsampling along columns.
% options.u = 5; % number of scales
% options.v = 8; % number of orientations
% options.mm = 39; % rows of filter bank
% options.nn = 39; % columns of filter bank
% options.show = 1; % display gabor masks
% I = imread('cameraman.tif'); % input image
% [X,Xn] = Bfx_gaborfull(I,[],options); % Gabor features
% Bio_printfeatures(X(1:10),Xn(1:10,:))
%
% See also Bfx_haralick, Bfx_clp, Bfx_fourier, Bfx_dct, Bfx_lbp, Bfx_gabor.
%
% (c) Functions gaborFilterBank and gaborFeatures were written by Mohammad Haghighat
% see credits below.
%
% (c) D.Mery, PUC-DCC, 2016
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_gaborfull(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
if isempty(R)
R = ones(size(I));
end
if options.show
disp('--- extracting Full Gabor features...');
end
A = gaborFilterBank(options); % Generates the Gabor filter bank
X = gaborFeatures(I,A,R,options); % Generates the Gabor features
n = numel(X);
Xn = char(zeros(n,24));
for i=1:n
Xn(i,:) = sprintf('Gab_%7d ',i);
end
function featureVector = gaborFeatures(img,gaborArray,R,options)
% GABORFEATURES extracts the Gabor features of an input image.
% It creates a column vector, consisting of the Gabor features of the input
% image. The feature vectors are normalized to zero mean and unit variance.
%
%
% Inputs:
% img : Matrix of the input image
% gaborArray : Gabor filters bank created by the function gaborFilterBank
% d1 : The factor of downsampling along rows.
% d2 : The factor of downsampling along columns.
%
% Output:
% featureVector : A column vector with length (m*n*u*v)/(d1*d2).
% This vector is the Gabor feature vector of an
% m by n image. u is the number of scales and
% v is the number of orientations in 'gaborArray'.
%
%
% Sample use:
%
% img = imread('cameraman.tif');
% gaborArray = gaborFilterBank(5,8,39,39); % Generates the Gabor filter bank
% featureVector = gaborFeatures(img,gaborArray,4,4); % Extracts Gabor feature vector, 'featureVector', from the image, 'img'.
%
%
%
% Details can be found in:
%
% M. Haghighat, S. Zonouz, M. Abdel-Mottaleb, "CloudID: Trustworthy
% cloud-based and cross-enterprise biometric identification,"
% Expert Systems with Applications, vol. 42, no. 21, pp. 7905-7916, 2015.
%
%
%
% (C) Mohammad Haghighat, University of Miami
% [email protected]
% PLEASE CITE THE ABOVE PAPER IF YOU USE THIS CODE.
if (nargin ~= 4) % Check correct number of arguments
error('Please use the correct number of input arguments!')
end
if size(img,3) == 3 % Check if the input image is grayscale
warning('The input RGB image is converted to grayscale!')
img = rgb2gray(img);
end
img = double(img).*R;
d1 = options.d1; % factor of downsampling along rows.
d2 = options.d2; % factor of downsampling along columns.
%% Filter the image using the Gabor filter bank
% Filter input image by each Gabor filter
[u,v] = size(gaborArray);
gaborResult = cell(u,v);
for i = 1:u
for j = 1:v
gaborResult{i,j} = imfilter(img, gaborArray{i,j});
end
end
%% Create feature vector
% Extract feature vector from input image
nn = numel(img)/d1/d2;
%featureVector = [];
featureVector = zeros(1,nn*u*v);
t = 0;
for i = 1:u
for j = 1:v
t = t+1;
gaborAbs = abs(gaborResult{i,j});
gaborAbs = downsample(gaborAbs,d1);
gaborAbs = downsample(gaborAbs.',d2);
gaborAbs = gaborAbs(:);
% Normalized to zero mean and unit variance. (if not applicable, please comment this line)
gaborAbs = (gaborAbs-mean(gaborAbs))/std(gaborAbs,1);
%featureVector = [featureVector; gaborAbs];
featureVector(1,indices(t,nn)) = gaborAbs;
end
end
%% Show filtered images (Please comment this section if not needed!)
% % Show real parts of Gabor-filtered images
% figure('NumberTitle','Off','Name','Real parts of Gabor filters');
% for i = 1:u
% for j = 1:v
% subplot(u,v,(i-1)*v+j)
% imshow(real(gaborResult{i,j}),[]);
% end
% end
%
% % Show magnitudes of Gabor-filtered images
% figure('NumberTitle','Off','Name','Magnitudes of Gabor filters');
% for i = 1:u
% for j = 1:v
% subplot(u,v,(i-1)*v+j)
% imshow(abs(gaborResult{i,j}),[]);
% end
% end
function gaborArray = gaborFilterBank(options)
% GABORFILTERBANK generates a custum Gabor filter bank.
% It creates a u by v cell array, whose elements are m by n matrices;
% each matrix being a 2-D Gabor filter.
%
%
% Inputs:
% u : No. of scales (usually set to 5)
% v : No. of orientations (usually set to 8)
% m : No. of rows in a 2-D Gabor filter (an odd integer number, usually set to 39)
% n : No. of columns in a 2-D Gabor filter (an odd integer number, usually set to 39)
%
% Output:
% gaborArray: A u by v array, element of which are m by n
% matries; each matrix being a 2-D Gabor filter
%
%
% Sample use:
%
% gaborArray = gaborFilterBank(5,8,39,39);
%
%
%
% Details can be found in:
%
% M. Haghighat, S. Zonouz, M. Abdel-Mottaleb, "CloudID: Trustworthy
% cloud-based and cross-enterprise biometric identification,"
% Expert Systems with Applications, vol. 42, no. 21, pp. 7905-7916, 2015.
%
%
%
% (C) Mohammad Haghighat, University of Miami
% [email protected]
% PLEASE CITE THE ABOVE PAPER IF YOU USE THIS CODE.
%if (nargin ~= 4) % Check correct number of arguments
% error('There must be four input arguments (Number of scales and orientations and the 2-D size of the filter)!')
%end
%% Create Gabor filters
% Create u*v gabor filters each being an m by n matrix
u = options.u; % number of scales
v = options.v; % number of orientations
m = options.mm; % rows of filter bank
n = options.nn; % columns of filter bank
gaborArray = cell(u,v);
fmax = 0.25;
gama = sqrt(2);
eta = sqrt(2);
for i = 1:u
fu = fmax/((sqrt(2))^(i-1));
alpha = fu/gama;
beta = fu/eta;
for j = 1:v
tetav = ((j-1)/v)*pi;
gFilter = zeros(m,n);
for x = 1:m
for y = 1:n
xprime = (x-((m+1)/2))*cos(tetav)+(y-((n+1)/2))*sin(tetav);
yprime = -(x-((m+1)/2))*sin(tetav)+(y-((n+1)/2))*cos(tetav);
gFilter(x,y) = (fu^2/(pi*gama*eta))*exp(-((alpha^2)*(xprime^2)+(beta^2)*(yprime^2)))*exp(1i*2*pi*fu*xprime);
end
end
gaborArray{i,j} = gFilter;
end
end
%% Show Gabor filters (Please comment this section if not needed!)
if options.show == 1
% Show magnitudes of Gabor filters:
figure('NumberTitle','Off','Name','Magnitudes of Gabor filters');
for i = 1:u
for j = 1:v
subplot(u,v,(i-1)*v+j);
imshow(abs(gaborArray{i,j}),[]);
end
end
% Show real parts of Gabor filters:
figure('NumberTitle','Off','Name','Real parts of Gabor filters');
for i = 1:u
for j = 1:v
subplot(u,v,(i-1)*v+j);
imshow(real(gaborArray{i,j}),[]);
end
end
end
|
github
|
domingomery/Balu-master
|
Bfx_lbp.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbp.m
| 18,701 |
utf_8
|
0b6d8f300debf9639dc1a587453b85dc
|
% [X,Xn,options] = Bfx_lbp(I,R,options)
% [X,Xn,options] = Bfx_lbp(I,options)
% [X,Xn] = Bfx_lbp(I,R,options)
% [X,Xn] = Bfx_lbp(I,options)
%
% Toolbox: Balu
% Local Binary Patterns features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% It calculates the LBP over the a regular grid of patches. The function
% uses Heikkila & Ahonen (see http://www.cse.oulu.fi/MVG/Research/LBP).
%
% It returns a matrix of uniform lbp82 descriptors for I, made by
% concatenating histograms of each grid cell in the image.
% Grid size is options.hdiv * options.vdiv
%
% R is a binary image or empty. If R is given the lbp will be computed
% the corresponding pixles R==0 in image I will be set to 0.
%
% Output:
% X is a matrix of size ((hdiv*vdiv) x 59), each row has a
% histogram corresponding to a grid cell. We use 59 bins.
% options.x of size hdiv*vdiv is the x coordinates of center of ith grid cell
% options.y of size hdiv*vdiv is the y coordinates of center of ith grid cell
% Both coordinates are calculated as if image was a square of side length 1.
%
% References:
% Ojala, T.; Pietikainen, M. & Maenpaa, T. Multiresolution gray-scale
% and rotation invariant texture classification with local binary
% patterns. IEEE Transactions on Pattern Analysis and Machine
% Intelligence, 2002, 24, 971-987.
%
% Mu, Y. et al (2008): Discriminative Local Binary Patterns for Human
% Detection in Personal Album. CVPR-2008.
%
% Example 1:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'u2'; % uniform LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% figure(2);bar(X) % histogram
%
% Example 2:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'ri'; % rotation-invariant LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% bar(X) % histogram
%
% Example 3:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 4:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 16; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 5:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.25; % angle sampling
% options.weight = 9; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % weighted LBP features
% bar(X) % histogram
% See also Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
% James Kapaldo updated this version for Matlab2014b
%
function [X,Xn,options] = Bfx_lbp(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
vdiv = options.vdiv;
hdiv = options.hdiv;
if ~isfield(options,'show')
options.show = 0;
end
if ~isfield(options,'normalize')
options.normalize = 0;
end
if options.show == 1
disp('--- extracting local binary patterns features...');
end
if ~isfield(options,'samples')
options.samples = 8;
end
if ~isfield(options,'integral')
options.integral = 0;
end
if ~isfield(options,'radius')
options.radius = log(options.samples)/log(2)-1;
end
if ~isfield(options,'semantic')
options.semantic = 0;
end
if ~isfield(options,'weight')
options.weight = 0;
end
LBPst = 'LBP';
if options.semantic>0
if ~isfield(options,'sk')
options.sk = 1;
end
mapping = getsmapping(options.samples,options.sk);
LBPst = ['s' LBPst];
st='8x8';
else
% mapping = getmapping(8,'u2');
if ~isfield(options,'mappingtype')
options.mappingtype = 'u2';
end
st = sprintf('%d,%s',options.samples,options.mappingtype);
mapping = getmapping(options.samples,options.mappingtype);
end
% get lbp image
if ~isempty(R);
I(R==0) = 0;
end
code_img = lbp(I,options.radius,options.samples,mapping,'');
[n1,n2] = size(code_img);
[N,M] = size(I);
Ilbp = zeros(size(I));
i1 = round((N-n1)/2);
j1 = round((M-n2)/2);
Ilbp(i1+1:i1+n1,j1+1:j1+n2) = code_img;
options.Ilbp = Ilbp;
if options.integral == 1
options.Hx = Bim_inthist(Ilbp+1,options.maxD);
end
ylen = round(n1/vdiv);
xlen = round(n2/hdiv);
% split image into blocks (saved as columns)
grid_img = im2col(code_img,[ylen, xlen], 'distinct');
if options.weight>0
LBPst = ['w' LBPst];
mt = 2*options.radius-1;
mt2 = mt^2;
Id = double(I);
switch options.weight
case 1
W = abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id);
case 2
W = (abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id))./(Id+1);
case 3
W = abs(medfilt2(Id,[mt mt])-Id);
case 4
W = abs(medfilt2(Id,[mt mt])-Id)./(Id+1);
case 5
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id);
case 6
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
case 7
Id = conv2(Id,ones(mt,mt)/mt2,'same');
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
case 8
Id = medfilt2(Id,[mt mt]);
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
case 9
Id = medfilt2(Id,[mt mt]);
W = abs(ordfilt2(Id,mt2-1,ones(mt,mt))-Id)./(Id+1);
otherwise
error('Bfx_lbp does not recognice options.weight = %d.',options.weight);
end
W = W(mt+1:end-mt,mt+1:end-mt);
grid_W = im2col(W,[ylen, xlen], 'distinct');
nwi = mapping.num;
nwj = size(grid_W,2);
nwk = size(grid_W,1);
desc = zeros(nwi,nwj);
for j=1:nwj
x = grid_img(:,j)+1;
y = grid_W(:,j);
d = zeros(nwi,1);
for k=1:nwk
d(x(k))=d(x(k))+y(k);
% d(x(k))=d(x(k))+1; % normal LBP each LBP has equal weight
end
desc(:,j) = d;
end
else
desc = hist(double(grid_img), 0:mapping.num-1);
% calculate coordinates of descriptors as if I was square w/ side=1
end
dx = 1.0/hdiv;
dy = 1.0/vdiv;
x = dx/2.0: dx :1.0-dx/2.0;
y = dy/2.0: dy :1.0-dy/2.0;
options.x = x;
options.y = y;
if hdiv*vdiv>1
D = desc';
else
D = desc;
end
[M,N] = size(D);
Xn = char(zeros(N*M,24));
X = zeros(1,N*M);
k=0;
for i=1:M
for j=1:N
k = k+1;
s = sprintf('%s(%d,%d)[%s] ',LBPst,i,j,st);
Xn(k,:) = s(1:24);
X(k) = D(i,j);
end
end
if options.normalize
X = X/sum(X);
end
end
%GETMAPPING returns a structure containing a mapping table for LBP codes.
% MAPPING = GETMAPPING(SAMPLES,MAPPINGTYPE) returns a
% structure containing a mapping table for
% LBP codes in a neighbourhood of SAMPLES sampling
% points. Possible values for MAPPINGTYPE are
% 'u2' for uniform LBP
% 'ri' for rotation-invariant LBP
% 'riu2' for uniform rotation-invariant LBP.
%
% Example:
% I=imread('rice.tif');
% MAPPING=getmapping(16,'riu2');
% LBPHIST=lbp(I,2,16,MAPPING,'hist');
% Now LBPHIST contains a rotation-invariant uniform LBP
% histogram in a (16,2) neighbourhood.
%
function mapping = getmapping(samples,mappingtype)
% Version 0.1.1
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% 0.1.1 Changed output to be a structure
% Fixed a bug causing out of memory errors when generating rotation
% invariant mappings with high number of sampling points.
% Lauge Sorensen is acknowledged for spotting this problem.
table = 0:2^samples-1;
newMax = 0; %number of patterns in the resulting LBP code
index = 0;
%vr2014b = or(strcmp(version('-release'),'2014b'),strcmp(version('-release'),'2014a'));
%if vr2014b
switch samples
case 8
sampleType = 'uint8';
case 16
sampleType = 'uint16';
otherwise
end
%else
% sampleType = samples;
%end
if strcmp(mappingtype,'u2') %Uniform 2
newMax = samples*(samples-1) + 3;
for i = 0:2^samples-1
j = bitset(bitshift(i,1,sampleType),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples)); %number of 1->0 and
%0->1 transitions
%in binary string
%x is equal to the
%number of 1-bits in
%XOR(x,Rotate left(x))
if numt <= 2
table(i+1) = index;
index = index + 1;
else
table(i+1) = newMax - 1;
end
end
end
if strcmp(mappingtype,'ri') %Rotation invariant
tmpMap = zeros(2^samples,1) - 1;
for i = 0:2^samples-1
rm = i;
r = i;
for j = 1:samples-1
r = bitset(bitshift(r,1,sampleType),1,bitget(r,samples)); %rotate
%left
if r < rm
rm = r;
end
end
if tmpMap(rm+1) < 0
tmpMap(rm+1) = newMax;
newMax = newMax + 1;
end
table(i+1) = tmpMap(rm+1);
end
end
if strcmp(mappingtype,'riu2') %Uniform & Rotation invariant
newMax = samples + 2;
for i = 0:2^samples - 1
j = bitset(bitshift(i,1,sampleType),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples));
if numt <= 2
table(i+1) = sum(bitget(i,1:samples));
else
table(i+1) = samples+1;
end
end
end
mapping.table=table;
mapping.samples=samples;
mapping.num=newMax;
end
% LBP returns the local binary pattern image or LBP histogram of an image.
% J = LBP(I,R,N,MAPPING,MODE) returns either a local binary pattern
% coded image or the local binary pattern histogram of an intensity
% image I. The LBP codes are computed using N sampling points on a
% circle of radius R and using mapping table defined by MAPPING.
% See the getmapping function for different mappings and use 0 for
% no mapping. Possible values for MODE are
% 'h' or 'hist' to get a histogram of LBP codes
% 'nh' to get a normalized histogram
% Otherwise an LBP code image is returned.
%
% J = LBP(I) returns the original (basic) LBP histogram of image I
%
% J = LBP(I,SP,MAPPING,MODE) computes the LBP codes using n sampling
% points defined in (n * 2) matrix SP. The sampling points should be
% defined around the origin (coordinates (0,0)).
%
% Examples
% --------
% I=imread('rice.png');
% mapping=getmapping(8,'u2');
% H1=LBP(I,1,8,mapping,'h'); %LBP histogram in (8,1) neighborhood
% %using uniform patterns
% subplot(2,1,1),stem(H1);
%
% H2=LBP(I);
% subplot(2,1,2),stem(H2);
%
% SP=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
% I2=LBP(I,SP,0,'i'); %LBP code image using sampling points in SP
% %and no mapping. Now H2 is equal to histogram
% %of I2.
function result = lbp(varargin) % image,radius,neighbors,mapping,mode)
% Version 0.3.2
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% Version 0.3.2: A bug fix to enable using mappings together with a
% predefined spoints array
% Version 0.3.1: Changed MAPPING input to be a struct containing the mapping
% table and the number of bins to make the function run faster with high number
% of sampling points. Lauge Sorensen is acknowledged for spotting this problem.
% Check number of input arguments.
narginchk(1,5);
% error(nargchk(1,5,nargin)); % for previous versions of matlab
image=varargin{1};
d_image=double(image);
if nargin==1
spoints=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
neighbors=8;
mapping=0;
mode='h';
end
if (nargin == 2) && (length(varargin{2}) == 1)
error('Input arguments');
end
if (nargin > 2) && (length(varargin{2}) == 1)
radius=varargin{2};
neighbors=varargin{3};
spoints=zeros(neighbors,2);
% Angle step.
a = 2*pi/neighbors;
for i = 1:neighbors
spoints(i,1) = -radius*sin((i-1)*a);
spoints(i,2) = radius*cos((i-1)*a);
end
if(nargin >= 4)
mapping=varargin{4};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 5)
mode=varargin{5};
else
mode='h';
end
end
if (nargin > 1) && (length(varargin{2}) > 1)
spoints=varargin{2};
neighbors=size(spoints,1);
if(nargin >= 3)
mapping=varargin{3};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 4)
mode=varargin{4};
else
mode='h';
end
end
% Determine the dimensions of the input image.
[ysize xsize] = size(image);
miny=min(spoints(:,1));
maxy=max(spoints(:,1));
minx=min(spoints(:,2));
maxx=max(spoints(:,2));
% Block size, each LBP code is computed within a block of size bsizey*bsizex
bsizey=ceil(max(maxy,0))-floor(min(miny,0))+1;
bsizex=ceil(max(maxx,0))-floor(min(minx,0))+1;
% Coordinates of origin (0,0) in the block
origy=1-floor(min(miny,0));
origx=1-floor(min(minx,0));
% Minimum allowed size for the input image depends
% on the radius of the used LBP operator.
if(xsize < bsizex || ysize < bsizey)
error('Too small input image. Should be at least (2*radius+1) x (2*radius+1)');
end
% Calculate dx and dy;
dx = xsize - bsizex;
dy = ysize - bsizey;
% Fill the center pixel matrix C.
C = image(origy:origy+dy,origx:origx+dx);
d_C = double(C);
bins = 2^neighbors;
% Initialize the result matrix with zeros.
result=zeros(dy+1,dx+1);
%Compute the LBP code image
for i = 1:neighbors
y = spoints(i,1)+origy;
x = spoints(i,2)+origx;
% Calculate floors, ceils and rounds for the x and y.
fy = floor(y); cy = ceil(y); ry = round(y);
fx = floor(x); cx = ceil(x); rx = round(x);
% Check if interpolation is needed.
if (abs(x - rx) < 1e-6) && (abs(y - ry) < 1e-6)
% Interpolation is not needed, use original datatypes
N = image(ry:ry+dy,rx:rx+dx);
D = N >= C;
else
% Interpolation needed, use double type images
ty = y - fy;
tx = x - fx;
% Calculate the interpolation weights.
w1 = (1 - tx) * (1 - ty);
w2 = tx * (1 - ty);
w3 = (1 - tx) * ty ;
w4 = tx * ty ;
% Compute interpolated pixel values
N = w1*d_image(fy:fy+dy,fx:fx+dx) + w2*d_image(fy:fy+dy,cx:cx+dx) + ...
w3*d_image(cy:cy+dy,fx:fx+dx) + w4*d_image(cy:cy+dy,cx:cx+dx);
D = N >= d_C;
end
% Update the result matrix.
v = 2^(i-1);
result = result + v*D;
end
%Apply mapping if it is defined
if isstruct(mapping)
bins = mapping.num;
for i = 1:size(result,1)
for j = 1:size(result,2)
result(i,j) = mapping.table(result(i,j)+1);
end
end
end
if (strcmp(mode,'h') || strcmp(mode,'hist') || strcmp(mode,'nh'))
% Return with LBP histogram if mode equals 'hist'.
result=hist(result(:),0:(bins-1));
if (strcmp(mode,'nh'))
result=result/sum(result);
end
else
%Otherwise return a matrix of unsigned integers
if ((bins-1)<=intmax('uint8'))
result=uint8(result);
elseif ((bins-1)<=intmax('uint16'))
result=uint16(result);
else
result=uint32(result);
end
end
end
% Mapping for sLBB
function mapping = getsmapping(N,sk)
vr2014b = or(strcmp(version('-release'),'2014b'),strcmp(version('-release'),'2014a'));
if vr2014b
switch N
case 8
sampleType = 'uint8';
case 16
sampleType = 'uint16';
otherwise
end
else
sampleType = samples;
end
M = 2^N;
samples = N;
len = zeros(M,1);
ang = zeros(M,1);
for x=0:(M-1)
k = x+1;
j = bitset(bitshift(x,1,sampleType),1,bitget(x,samples));
numt = sum(bitget(bitxor(x,j),1:samples));
c = numt;
if c>2
len(k)=-1;
ang(k)=-1;
else
s = bitget(x,1:samples);
len(k) = sum(s);
if c==0
ang(k)=0;
else
r = 0;
while (s(1)~=0) || (s(N)~=1)
s = [s(2:N) s(1)];
r = r+1;
end
ii = find(s==1);
a = mean(ii)+r;
if a>N
a=a-N;
end
ang(k) = round(sk*a)-1;
end
end
% fprintf('%4d: %s (%d,%d)\n',x,dec2bin(x,N),len(k),ang(k)); pause
end
Ma = max(ang)+1;
map = len*Ma+ang-Ma+1;
n = max(map)+1;
map(ang==-1) = n;
map(1) = 0;
mapping.table = map';
mapping.samples = N;
mapping.num = n+1;
end
|
github
|
domingomery/Balu-master
|
Bfx_lbpcontrast.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbpcontrast.m
| 1,609 |
utf_8
|
44e9e37cce48e0b0ab58dbb5c1aa27f0
|
% Pietikainen, M. et al (2011): Computer Vision Using Local Binary
% Patterns, Springer.
function J = Bfx_lbpcontrast(I,options)
if ~exist('options','var')
m = 3;
else
m = options.m;
end
n = (m-1)/2;
[N,M,P] = size(I);
if P==1 % 2D
m0 = (m^2+1)/2;
ii = [1:m0-1 m0+1:m^2];
J = zeros(N,M);
for i=n+1:N-n
for j=n+1:M-n
s = I(i-n:i+n,j-n:j+n);
if std2(s)>0
st = s(ii);
it = st>=s(m0);
Jplus = mean(st(it));
Jminus = mean(st(not(it)));
if isnan(Jplus)
Jplus = 0;
Jminus = 0;
end
if isnan(Jminus)
Jminus = 0;
Jplus = 0;
end
J(i,j) = Jplus-Jminus;
end
end
end
else
m0 = (m^3+1)/2;
ii = [1:m0-1 m0+1:m^3];
J = zeros(N,M,P);
for i=n+1:N-1
for j=n+1:M-1
for k=n+1:P-1
s = I(i-n:i+n,j-n:j+n,k-n:k+n);
if std2(s)>0
st = s(ii);
it = st>=s(m0);
Jplus = mean(st(it));
Jminus = mean(st(not(it)));
if isnan(Jplus)
Jplus = 0;
Jminus = 0;
end
if isnan(Jminus)
Jminus = 0;
Jplus = 0;
end
J(i,j,k) = Jplus-Jminus;
end
end
end
end
end
|
github
|
domingomery/Balu-master
|
Bfx_flusser.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_flusser.m
| 2,208 |
utf_8
|
05663d5a0b328c25995959249565bdf2
|
% [X,Xn] = Bfx_flusser(R,options)
% [X,Xn] = Bfx_flusser(R)
%
% Toolbox: Balu
%
% Extract the four Flusser moments from binary image R.
%
% options.show = 1 display mesagges.
%
% X is a 4 elements vector:
% X(i): Flusser-moment i for i=1,...,4.
% Xn is the list of feature names.
%
% Reference:
% Sonka et al. (1998): Image Processing, Analysis, and Machine Vision,
% PWS Publishing. Pacific Grove, Ca, 2nd Edition.
%
% Example:
% I = imread('testimg3.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [L,n] = bwlabel(R); % regions
% imshow(L,[])
% X = [];
% for i=1:n
% [Xi,Xn] = Bfx_flusser(L==i); % Flusser moments
% X = [X;Xi];
% end
% X
%
% See also Bfx_standard, Bfx_hugeo, Bfx_fitellipse, Bfx_gupta.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_flusser(R,options)
if ~exist('options','var')
options.show = 0;
end
if options.show == 1
disp('--- extracting Flusser moments...');
end
[Ireg,Jreg] = find(R==1); % pixels in the region
i_m = mean(Ireg);
j_m = mean(Jreg);
A = length(Ireg);
I0 = ones(A,1);
J0 = ones(A,1);
I1 = Ireg - i_m*ones(A,1);
J1 = Jreg - j_m*ones(A,1);
I2 = I1.*I1;
J2 = J1.*J1;
I3 = I2.*I1;
J3 = J2.*J1;
% Central moments
u00 = (I0'*J0);
% u01 = (I0'*J1); not used
u02 = (I0'*J2);
u03 = (I0'*J3);
% u10 = (I1'*J0); not used
u20 = (I2'*J0);
u30 = (I3'*J0);
u11 = (I1'*J1);
u12 = (I1'*J2);
u21 = (I2'*J1);
II1 = (u20*u02-u11^2)/u00^4 ;
II2 = (u30^2*u03^2-6*u30*u21*u12*u03+4*u30*u12^3+4*u21^3*u03-3*u21^2*u12^2)/u00^10;
II3 = (u20*(u21*u03-u12^2)-u11*(u30*u03-u21*u12)+u02*(u30*u12-u21^2))/u00^7;
II4 = (u20^3*u03^2-6*u20^2*u11*u12*u03-6*u20^2*u02*u21*u03+9*u20^2*u02*u12^2 + 12*u20*u11^2*u21*u03+6*u20*u11*u02*u30*u03-18*u20*u11*u02*u21*u12-8*u11^3*u30*u03- 6*u20*u02^2*u30*u12+9*u20*u02^2*u21+12*u11^2*u02*u30*u12-6*u11*u02^2*u30*u21+u02^3*u30^2)/u00^11;
X = [II1 II2 II3 II4];
Xn = [ 'Flusser-moment 1 '
'Flusser-moment 2 '
'Flusser-moment 3 '
'Flusser-moment 4 '];
|
github
|
domingomery/Balu-master
|
Bfx_phog.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_phog.m
| 4,705 |
utf_8
|
76718a146565c533700f9b922c118f81
|
% [X,Xn,Xu] = Bfx_phog(I,R,options)
% [X,Xn,Xu] = Bfx_phog(I,options)
%
% Toolbox: Balu
%
% Pyramid Histogram of Oriented Gradients based on implementation by
% Anna Bosch from
%
% http://www.robots.ox.ac.uk/~vgg/research/caltech/phog.html
%
% IN:
% I - Images of size MxN (Color or Gray)
% options.bin - Number of bins on the histogram
% options.L - number of pyramid levels
%
% OUT:
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% Reference:
% Dalal, N. & Triggs, B. (2005): Histograms of oriented gradients for
% human detection, Proceedings of the Conference on Computer Vision and
% Pattern Recognition, Vol. 1, 886-893
%
% Example:
% options.bin = 9; % bins on the histogram
% options.L = 3; % pyramides levels
% options.show = 1; % display results
% I = imread('testimg1.jpg'); % input image
% J = double(I(:,:,2))/256; % normalized green channel
% [X,Xn] = Bfx_phog(J,options); % phog features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_haralick, Bfx_clp, Bfx_gabor, Bfx_fourier, Bfx_lbp.
%
% D.Mery, A. Soto PUC-DCC, Jun. 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_phog(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
I(R==0) = 0;
bin = options.bin;
L = options.L;
if options.show
disp('--- extracting phog features...');
end
roi = [1 size(I,1) 1 size(I,2)]';
angle = 360;
if size(I,3) == 3
G = rgb2gray(I);
else
G = I;
end
if sum(sum(G))>100
E = edge(G,'canny');
[GradientX,GradientY] = gradient(double(G));
Gr = sqrt((GradientX.*GradientX)+(GradientY.*GradientY));
index = GradientX == 0;
GradientX(index) = 1e-5;
A = ((atan2(GradientY,GradientX)+pi)*180)/pi;
[bh bv] = BphogbinMatrix(A,E,Gr,angle,bin);
else
bh = zeros(size(I,1),size(I,2));
bv = zeros(size(I,1),size(I,2));
end
bh_roi = bh(roi(1,1):roi(2,1),roi(3,1):roi(4,1));
bv_roi = bv(roi(1,1):roi(2,1),roi(3,1):roi(4,1));
X = BphogDescriptor(bh_roi,bv_roi,L,bin)';
n = length(X);
Xn = char(zeros(n,24));
for i=1:n
s = sprintf('phog(%d) ',i);
Xn(i,:) = s(1:24);
end
end
function p = BphogDescriptor(bh,bv,L,bin)
% anna_PHOGDESCRIPTOR Computes Pyramid Histogram of Oriented Gradient over a ROI.
%
% Pyramid Histogram of Oriented Gradients based on implementation by
% Anna Bosch from
%
% http://www.robots.ox.ac.uk/~vgg/research/caltech/phog.html
%
%IN:
% bh - matrix of bin histogram values
% bv - matrix of gradient values
% L - number of pyramid levels
% bin - number of bins
%
%OUT:
% p - pyramid histogram of oriented gradients (phog descriptor)
p = [];
for b=1:bin
ind = bh==b;
p = [p;sum(bv(ind))];
end
cella = 1;
for l=1:L
x = fix(size(bh,2)/(2^l));
y = fix(size(bh,1)/(2^l));
xx=0;
yy=0;
while xx+x<=size(bh,2)
while yy +y <=size(bh,1)
bh_cella = bh(yy+1:yy+y,xx+1:xx+x);
bv_cella = bv(yy+1:yy+y,xx+1:xx+x);
for b=1:bin
ind = bh_cella==b;
p = [p;sum(bv_cella(ind))];
end
yy = yy+y;
end
cella = cella+1;
yy = 0;
xx = xx+x;
end
end
if sum(p)~=0
p = p/sum(p);
end
end
function [bm bv] = BphogbinMatrix(A,E,G,angle,bin)
% anna_BINMATRIX Computes a Matrix (bm) with the same size of the image where
% (i,j) position contains the histogram value for the pixel at position (i,j)
% and another matrix (bv) where the position (i,j) contains the gradient
% value for the pixel at position (i,j)
%
% Pyramid Histogram of Oriented Gradients based on implementation by
% Anna Bosch from
%
% http://www.robots.ox.ac.uk/~vgg/research/caltech/phog.html
%
%
%IN:
% A - Matrix containing the angle values
% E - Edge Image
% G - Matrix containing the gradient values
% angle - 180 or 360%
% bin - Number of bins on the histogram
% angle - 180 or 360
%OUT:
% bm - matrix with the histogram values
% bv - matrix with the graident values (only for the pixels belonging to
% and edge)
[contorns,n] = bwlabel(E);
X = size(E,2);
Y = size(E,1);
bm = zeros(Y,X);
bv = zeros(Y,X);
nAngle = angle/bin;
for i=1:n
[posY,posX] = find(contorns==i);
for j=1:size(posY,1)
pos_x = posX(j,1);
pos_y = posY(j,1);
b = ceil(A(pos_y,pos_x)/nAngle);
if b==0, bin= 1; end
if G(pos_y,pos_x)>0
bm(pos_y,pos_x) = b;
bv(pos_y,pos_x) = G(pos_y,pos_x);
end
end
end
end
|
github
|
domingomery/Balu-master
|
Bfx_clp_old.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_clp_old.m
| 4,664 |
utf_8
|
c8bae5401c7b894bb426467bef73f207
|
% [X,Xn] = Bfx_clp(I,R,options)
% [X,Xn] = Bfx_clp(I,options)
%
% Toolbox: Balu
% Crossing Line Profile.
%
% X is the features vector, Xn is the list of feature names(see Example
% to see how it works).
%
% Reference:
% Mery, D.: Crossing line profile: a new approach to detecting defects
% in aluminium castings. Proceedings of the Scandinavian Conference on
% Image Analysis 2003 (SCIA 2003), Lecture Notes in Computer Science
% LNCS 2749: 725-732, 2003.
%
% Example:
% options.show = 1; % display results
% options.ng = 32; % windows resize
% I = imread('testimg4.jpg'); % input image
% J = I(395:425,415:442,1); % region of interest (red)
% R = J>135; % segmentation
% figure;imshow(J,[])
% figure;imshow(R)
% [X,Xn] = Bfx_clp(J,R,options); % CLP features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_contrast, Bfx_haralick, Bfx_clp, Bfx_fourier, Bfx_dct, Bfx_lbp.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_clp(I,R,options)
I = double(I);
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'show')
options.show = 0;
end
if options.show == 1
disp('--- extracting Crossing line profile features...');
end
show = options.show;
if ~isempty(R);
I(R==0) = 0;
end
ng = options.ng;
Bn = imresize(I,[ng ng]);
mg = fix(ng/2+1);
% Crossing line profiles
P0 = Bn(mg,:)'; % 0.0 ... 90.0 4
P1 = Bn(:,mg); % 90.0 ... 0.0 0
P2 = zeros(ng,1); % 45.0 ... 135.0 6
P3 = zeros(ng,1); % 135.0 ... 45.0 2
P4 = zeros(ng,1); % 22.5 ... 112.5 5
P5 = zeros(ng,1); % 67.5 ... 157.5 7
P6 = zeros(ng,1); % 112.5 ... 22.5 1
P7 = zeros(ng,1); % 157.5 ... 67.5 3
Q0 = Bn; Q1 = Bn; Q2 = Bn; Q3 = Bn;
Q4 = Bn; Q5 = Bn; Q6 = Bn; Q7 = Bn;
Q0(mg,:) = 255*ones(1,ng);
Q1(:,mg) = 255*ones(ng,1);
m4 = mg/(ng-1);
b4 = mg/2-m4;
b7 = 3*mg/2+mg/(ng-1);
for i=1:ng
P2(i,1) = Bn(i,i);
Q2(i,i) = 255;
P3(i,1) = Bn(i,ng-i+1);
Q3(i,ng-i+1) = 255;
j4 = fix(m4*i + b4 + 0.5);
P4(i,1) = Bn(j4,i);
Q4(j4,i) = 255;
P5(i,1) = Bn(i,j4);
Q5(i,j4) = 255;
j7 = fix(-m4*i + b7 + 0.5);
P6(i,1) = Bn(i,j7);
Q6(i,j7) = 255;
P7(i,1) = Bn(j7,i);
Q7(j7,i) = 255;
end
PP = [P0 P1 P2 P3 P4 P5 P6 P7];
d = abs(PP(1,:)-PP(ng,:));
[I,J] = sort(d);
Po = PP(:,J(1));
Po = Po/Po(1);
m = (Po(ng)-Po(1))/(ng - 1);
mb = Po(1)-m;
Q = Po-(1:ng)'*m-ones(ng,1)*mb;
Qm = mean(Q);
Qd = max(Q)-min(Q);
Qd1 = log(Qd+1);
Qd2 = 2*Qd/(Po(1)+Po(ng));
Qs = std(Q);
Qf = fft(Q);
Qf = abs(Qf(2:8,1));
if (show)
figure(10)
clf
subplot(2,4,1);plot(PP(:,1));axis([1 ng 0 255]);title('k=4');
subplot(2,4,2);plot(PP(:,2));axis([1 ng 0 255]);title('k=0');
subplot(2,4,3);plot(PP(:,3));axis([1 ng 0 255]);title('k=2');
subplot(2,4,4);plot(PP(:,4));axis([1 ng 0 255]);title('k=6');
subplot(2,4,5);plot(PP(:,5));axis([1 ng 0 255]);title('k=3');
subplot(2,4,6);plot(PP(:,6));axis([1 ng 0 255]);title('k=1');
subplot(2,4,7);plot(PP(:,7));axis([1 ng 0 255]);title('k=7');
subplot(2,4,8);plot(PP(:,8));axis([1 ng 0 255]);title('k=5');
figure(11)
clf
subplot(2,4,5);plot(PP(:,1));axis([1 ng 0 255]);title('k=4');
subplot(2,4,1);plot(PP(:,2));axis([1 ng 0 255]);title('k=0');
subplot(2,4,3);plot(PP(:,3));axis([1 ng 0 255]);title('k=2');
subplot(2,4,7);plot(PP(:,4));axis([1 ng 0 255]);title('k=6');
subplot(2,4,4);plot(PP(:,5));axis([1 ng 0 255]);title('k=3');
subplot(2,4,2);plot(PP(:,6));axis([1 ng 0 255]);title('k=1');
subplot(2,4,8);plot(PP(:,7));axis([1 ng 0 255]);title('k=7');
subplot(2,4,6);plot(PP(:,8));axis([1 ng 0 255]);title('k=5');
figure(12)
imshow([Q0 Q1 Q2 Q3;Q4 Q5 Q6 Q7],gray(256));
figure(13)
imshow([Q1 Q5 Q2 Q4;Q0 Q7 Q3 Q6],gray(256));
pause(0);
end
X = [Qm Qs Qd Qd1 Qd2 Qf'];
Xn = [ 'CLP-Qm '
'CLP-Qs '
'CLP-Qd '
'CLP-Qd1 '
'CLP-Qd2 '
'CLP-Qf1 '
'CLP-Qf2 '
'CLP-Qf3 '
'CLP-Qf4 '
'CLP-Qf5 '
'CLP-Qf6 '
'CLP-Qf7 '];
|
github
|
domingomery/Balu-master
|
Bfx_lbpi.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbpi.m
| 17,985 |
utf_8
|
41dcbf073502e4149f5abf197cc3092e
|
% [X,Xn,options] = Bfx_lbp(I,R,options)
% [X,Xn,options] = Bfx_lbp(I,options)
% [X,Xn] = Bfx_lbp(I,R,options)
% [X,Xn] = Bfx_lbp(I,options)
%
% Toolbox: Balu
% Local Binary Patterns features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% It calculates the LBP over the a regular grid of patches. The function
% uses Heikkila & Ahonen (see http://www.cse.oulu.fi/MVG/Research/LBP).
%
% It returns a matrix of uniform lbp82 descriptors for I, made by
% concatenating histograms of each grid cell in the image.
% Grid size is options.hdiv * options.vdiv
%
% R is a binary image or empty. If R is given the lbp will be computed
% the corresponding pixles R==0 in image I will be set to 0.
%
% Output:
% X is a matrix of size ((hdiv*vdiv) x 59), each row has a
% histogram corresponding to a grid cell. We use 59 bins.
% options.x of size hdiv*vdiv is the x coordinates of center of ith grid cell
% options.y of size hdiv*vdiv is the y coordinates of center of ith grid cell
% Both coordinates are calculated as if image was a square of side length 1.
%
% References:
% Ojala, T.; Pietikainen, M. & Maenpaa, T. Multiresolution gray-scale
% and rotation invariant texture classification with local binary
% patterns. IEEE Transactions on Pattern Analysis and Machine
% Intelligence, 2002, 24, 971-987.
%
% Mu, Y. et al (2008): Discriminative Local Binary Patterns for Human
% Detection in Personal Album. CVPR-2008.
%
% Example 1:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'u2'; % uniform LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% bar(X) % histogram
%
% Example 2:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'ri'; % rotation-invariant LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% bar(X) % histogram
%
% Example 3:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 4:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 16; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 5:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.25; % angle sampling
% options.weight = 9; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % weighted LBP features
% bar(X) % histogram
% See also Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_lbpi(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
if ~isfield(options,'normalize')
options.normalize = 0;
end
% vdiv = options.vdiv;
% hdiv = options.hdiv;
%
% if ~isfield(options,'show')
% options.show = 0;
% end
%
% if options.show == 1
% disp('--- extracting local binary patterns features...');
% end
%
% if ~isfield(options,'samples')
% options.samples = 8;
% end
%
% if ~isfield(options,'radius')
% options.radius = log(options.samples)/log(2)-1;
% end
%
% if ~isfield(options,'semantic')
% options.semantic = 0;
% end
%
% if ~isfield(options,'weight')
% options.weight = 0;
% end
LBPst = 'LBP';
st = 'i';
% if options.semantic>0
% if ~isfield(options,'sk')
% options.sk = 1;
% end
% mapping = getsmapping(options.samples,options.sk);
% LBPst = ['s' LBPst];
% st='8x8';
% else
% % mapping = getmapping(8,'u2');
% if ~isfield(options,'mappingtype')
% options.mappingtype = 'u2';
% end
% st = sprintf('%d,%s',options.samples,options.mappingtype);
% mapping = getmapping(options.samples,options.mappingtype);
% end
% get lbp image
if ~isempty(R);
I(R==0) = 0;
end
code_img = I;
[n1,n2] = size(code_img);
% [N,M] = size(I);
% Ilbp = zeros(size(I));
% i1 = round((N-n1)/2);
% j1 = round((M-n2)/2);
% code_img = Ilbp(i1+1:i1+n1,j1+1:j1+n2);
% options.Ilbp = Ilbp;
%ylen = round(n1/vdiv);
%xlen = round(n2/hdiv);
% split image into blocks (saved as columns)
%grid_img = im2col(code_img,[ylen, xlen], 'distinct');
grid_img = code_img(:);
% if options.weight>0
% LBPst = ['w' LBPst];
% mt = 2*options.radius-1;
% mt2 = mt^2;
% Id = double(I);
% switch options.weight
% case 1
% W = abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id);
% case 2
% W = (abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id))./(Id+1);
% case 3
% W = abs(medfilt2(Id,[mt mt])-Id);
% case 4
% W = abs(medfilt2(Id,[mt mt])-Id)./(Id+1);
% case 5
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id);
% case 6
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
% case 7
% Id = conv2(Id,ones(mt,mt)/mt2,'same');
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
% case 8
% Id = medfilt2(Id,[mt mt]);
% W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
% case 9
% Id = medfilt2(Id,[mt mt]);
% W = abs(ordfilt2(Id,mt2-1,ones(mt,mt))-Id)./(Id+1);
% otherwise
% error('Bfx_lbp does not recognice options.weight = %d.',options.weight);
% end
% W = W(mt+1:end-mt,mt+1:end-mt);
% grid_W = im2col(W,[ylen, xlen], 'distinct');
% nwi = mapping.num;
% nwj = size(grid_W,2);
% nwk = size(grid_W,1);
% desc = zeros(nwi,nwj);
% for j=1:nwj
% x = grid_img(:,j)+1;
% y = grid_W(:,j);
% d = zeros(nwi,1);
% for k=1:nwk
% d(x(k))=d(x(k))+y(k);
% % d(x(k))=d(x(k))+1; % normal LBP each LBP has equal weight
% end
% desc(:,j) = d;
% end
%
% else
desc = hist(double(grid_img), 0:options.maxD-1);
% calculate coordinates of descriptors as if I was square w/ side=1
%end
%dx = 1.0/hdiv;
%dy = 1.0/vdiv;
dx = 1;
dy = 1;
x = dx/2.0: dx :1.0-dx/2.0;
y = dy/2.0: dy :1.0-dy/2.0;
options.x = x;
options.y = y;
%if hdiv*vdiv>1
% D = desc';
%else
X = desc;
if options.normalize
X = X/sum(X);
end
%end
[M,N] = size(X);
Xn = char(zeros(N*M,24));
% X = zeros(1,N*M);
% k=0;
% for i=1:M
% for j=1:N
% k = k+1;
% s = sprintf('%s(%d,%d)[%s] ',LBPst,i,j,st);
% Xn(k,:) = s(1:24);
% X(k) = D(i,j);
% end
% end
end
%GETMAPPING returns a structure containing a mapping table for LBP codes.
% MAPPING = GETMAPPING(SAMPLES,MAPPINGTYPE) returns a
% structure containing a mapping table for
% LBP codes in a neighbourhood of SAMPLES sampling
% points. Possible values for MAPPINGTYPE are
% 'u2' for uniform LBP
% 'ri' for rotation-invariant LBP
% 'riu2' for uniform rotation-invariant LBP.
%
% Example:
% I=imread('rice.tif');
% MAPPING=getmapping(16,'riu2');
% LBPHIST=lbp(I,2,16,MAPPING,'hist');
% Now LBPHIST contains a rotation-invariant uniform LBP
% histogram in a (16,2) neighbourhood.
%
function mapping = getmapping(samples,mappingtype)
% Version 0.1.1
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% 0.1.1 Changed output to be a structure
% Fixed a bug causing out of memory errors when generating rotation
% invariant mappings with high number of sampling points.
% Lauge Sorensen is acknowledged for spotting this problem.
table = 0:2^samples-1;
newMax = 0; %number of patterns in the resulting LBP code
index = 0;
if strcmp(mappingtype,'u2') %Uniform 2
newMax = samples*(samples-1) + 3;
for i = 0:2^samples-1
j = bitset(bitshift(i,1,samples),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples)); %number of 1->0 and
%0->1 transitions
%in binary string
%x is equal to the
%number of 1-bits in
%XOR(x,Rotate left(x))
if numt <= 2
table(i+1) = index;
index = index + 1;
else
table(i+1) = newMax - 1;
end
end
end
if strcmp(mappingtype,'ri') %Rotation invariant
tmpMap = zeros(2^samples,1) - 1;
for i = 0:2^samples-1
rm = i;
r = i;
for j = 1:samples-1
r = bitset(bitshift(r,1,samples),1,bitget(r,samples)); %rotate
%left
if r < rm
rm = r;
end
end
if tmpMap(rm+1) < 0
tmpMap(rm+1) = newMax;
newMax = newMax + 1;
end
table(i+1) = tmpMap(rm+1);
end
end
if strcmp(mappingtype,'riu2') %Uniform & Rotation invariant
newMax = samples + 2;
for i = 0:2^samples - 1
j = bitset(bitshift(i,1,samples),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples));
if numt <= 2
table(i+1) = sum(bitget(i,1:samples));
else
table(i+1) = samples+1;
end
end
end
mapping.table=table;
mapping.samples=samples;
mapping.num=newMax;
end
% LBP returns the local binary pattern image or LBP histogram of an image.
% J = LBP(I,R,N,MAPPING,MODE) returns either a local binary pattern
% coded image or the local binary pattern histogram of an intensity
% image I. The LBP codes are computed using N sampling points on a
% circle of radius R and using mapping table defined by MAPPING.
% See the getmapping function for different mappings and use 0 for
% no mapping. Possible values for MODE are
% 'h' or 'hist' to get a histogram of LBP codes
% 'nh' to get a normalized histogram
% Otherwise an LBP code image is returned.
%
% J = LBP(I) returns the original (basic) LBP histogram of image I
%
% J = LBP(I,SP,MAPPING,MODE) computes the LBP codes using n sampling
% points defined in (n * 2) matrix SP. The sampling points should be
% defined around the origin (coordinates (0,0)).
%
% Examples
% --------
% I=imread('rice.png');
% mapping=getmapping(8,'u2');
% H1=LBP(I,1,8,mapping,'h'); %LBP histogram in (8,1) neighborhood
% %using uniform patterns
% subplot(2,1,1),stem(H1);
%
% H2=LBP(I);
% subplot(2,1,2),stem(H2);
%
% SP=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
% I2=LBP(I,SP,0,'i'); %LBP code image using sampling points in SP
% %and no mapping. Now H2 is equal to histogram
% %of I2.
function result = lbp(varargin) % image,radius,neighbors,mapping,mode)
% Version 0.3.2
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% Version 0.3.2: A bug fix to enable using mappings together with a
% predefined spoints array
% Version 0.3.1: Changed MAPPING input to be a struct containing the mapping
% table and the number of bins to make the function run faster with high number
% of sampling points. Lauge Sorensen is acknowledged for spotting this problem.
% Check number of input arguments.
error(nargchk(1,5,nargin));
image=varargin{1};
d_image=double(image);
if nargin==1
spoints=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
neighbors=8;
mapping=0;
mode='h';
end
if (nargin == 2) && (length(varargin{2}) == 1)
error('Input arguments');
end
if (nargin > 2) && (length(varargin{2}) == 1)
radius=varargin{2};
neighbors=varargin{3};
spoints=zeros(neighbors,2);
% Angle step.
a = 2*pi/neighbors;
for i = 1:neighbors
spoints(i,1) = -radius*sin((i-1)*a);
spoints(i,2) = radius*cos((i-1)*a);
end
if(nargin >= 4)
mapping=varargin{4};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 5)
mode=varargin{5};
else
mode='h';
end
end
if (nargin > 1) && (length(varargin{2}) > 1)
spoints=varargin{2};
neighbors=size(spoints,1);
if(nargin >= 3)
mapping=varargin{3};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 4)
mode=varargin{4};
else
mode='h';
end
end
% Determine the dimensions of the input image.
[ysize xsize] = size(image);
miny=min(spoints(:,1));
maxy=max(spoints(:,1));
minx=min(spoints(:,2));
maxx=max(spoints(:,2));
% Block size, each LBP code is computed within a block of size bsizey*bsizex
bsizey=ceil(max(maxy,0))-floor(min(miny,0))+1;
bsizex=ceil(max(maxx,0))-floor(min(minx,0))+1;
% Coordinates of origin (0,0) in the block
origy=1-floor(min(miny,0));
origx=1-floor(min(minx,0));
% Minimum allowed size for the input image depends
% on the radius of the used LBP operator.
if(xsize < bsizex || ysize < bsizey)
error('Too small input image. Should be at least (2*radius+1) x (2*radius+1)');
end
% Calculate dx and dy;
dx = xsize - bsizex;
dy = ysize - bsizey;
% Fill the center pixel matrix C.
C = image(origy:origy+dy,origx:origx+dx);
d_C = double(C);
bins = 2^neighbors;
% Initialize the result matrix with zeros.
result=zeros(dy+1,dx+1);
%Compute the LBP code image
for i = 1:neighbors
y = spoints(i,1)+origy;
x = spoints(i,2)+origx;
% Calculate floors, ceils and rounds for the x and y.
fy = floor(y); cy = ceil(y); ry = round(y);
fx = floor(x); cx = ceil(x); rx = round(x);
% Check if interpolation is needed.
if (abs(x - rx) < 1e-6) && (abs(y - ry) < 1e-6)
% Interpolation is not needed, use original datatypes
N = image(ry:ry+dy,rx:rx+dx);
D = N >= C;
else
% Interpolation needed, use double type images
ty = y - fy;
tx = x - fx;
% Calculate the interpolation weights.
w1 = (1 - tx) * (1 - ty);
w2 = tx * (1 - ty);
w3 = (1 - tx) * ty ;
w4 = tx * ty ;
% Compute interpolated pixel values
N = w1*d_image(fy:fy+dy,fx:fx+dx) + w2*d_image(fy:fy+dy,cx:cx+dx) + ...
w3*d_image(cy:cy+dy,fx:fx+dx) + w4*d_image(cy:cy+dy,cx:cx+dx);
D = N >= d_C;
end
% Update the result matrix.
v = 2^(i-1);
result = result + v*D;
end
%Apply mapping if it is defined
if isstruct(mapping)
bins = mapping.num;
for i = 1:size(result,1)
for j = 1:size(result,2)
result(i,j) = mapping.table(result(i,j)+1);
end
end
end
if (strcmp(mode,'h') || strcmp(mode,'hist') || strcmp(mode,'nh'))
% Return with LBP histogram if mode equals 'hist'.
result=hist(result(:),0:(bins-1));
if (strcmp(mode,'nh'))
result=result/sum(result);
end
else
%Otherwise return a matrix of unsigned integers
if ((bins-1)<=intmax('uint8'))
result=uint8(result);
elseif ((bins-1)<=intmax('uint16'))
result=uint16(result);
else
result=uint32(result);
end
end
end
% Mapping for sLBB
function mapping = getsmapping(N,sk)
M = 2^N;
samples = N;
len = zeros(M,1);
ang = zeros(M,1);
for x=0:(M-1)
k = x+1;
j = bitset(bitshift(x,1,samples),1,bitget(x,samples));
numt = sum(bitget(bitxor(x,j),1:samples));
c = numt;
if c>2
len(k)=-1;
ang(k)=-1;
else
s = bitget(x,1:samples);
len(k) = sum(s);
if c==0
ang(k)=0;
else
r = 0;
while (s(1)~=0) || (s(N)~=1)
s = [s(2:N) s(1)];
r = r+1;
end
ii = find(s==1);
a = mean(ii)+r;
if a>N
a=a-N;
end
ang(k) = round(sk*a)-1;
end
end
% fprintf('%4d: %s (%d,%d)\n',x,dec2bin(x,N),len(k),ang(k)); pause
end
Ma = max(ang)+1;
map = len*Ma+ang-Ma+1;
n = max(map)+1;
map(ang==-1) = n;
map(1) = 0;
mapping.table = map';
mapping.samples = N;
mapping.num = n+1;
end
|
github
|
domingomery/Balu-master
|
Bfx_lbp_old.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_lbp_old.m
| 18,020 |
utf_8
|
5d0a1fbb2228d5669c26283336b539fe
|
% [X,Xn,options] = Bfx_lbp(I,R,options)
% [X,Xn,options] = Bfx_lbp(I,options)
% [X,Xn] = Bfx_lbp(I,R,options)
% [X,Xn] = Bfx_lbp(I,options)
%
% Toolbox: Balu
% Local Binary Patterns features
%
% X is the features vector, Xn is the list of feature names (see Example
% to see how it works).
%
% It calculates the LBP over the a regular grid of patches. The function
% uses Heikkila & Ahonen (see http://www.cse.oulu.fi/MVG/Research/LBP).
%
% It returns a matrix of uniform lbp82 descriptors for I, made by
% concatenating histograms of each grid cell in the image.
% Grid size is options.hdiv * options.vdiv
%
% R is a binary image or empty. If R is given the lbp will be computed
% the corresponding pixles R==0 in image I will be set to 0.
%
% Output:
% X is a matrix of size ((hdiv*vdiv) x 59), each row has a
% histogram corresponding to a grid cell. We use 59 bins.
% options.x of size hdiv*vdiv is the x coordinates of center of ith grid cell
% options.y of size hdiv*vdiv is the y coordinates of center of ith grid cell
% Both coordinates are calculated as if image was a square of side length 1.
%
% References:
% Ojala, T.; Pietikainen, M. & Maenpaa, T. Multiresolution gray-scale
% and rotation invariant texture classification with local binary
% patterns. IEEE Transactions on Pattern Analysis and Machine
% Intelligence, 2002, 24, 971-987.
%
% Mu, Y. et al (2008): Discriminative Local Binary Patterns for Human
% Detection in Personal Album. CVPR-2008.
%
% Example 1:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'u2'; % uniform LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% figure(2);bar(X) % histogram
%
% Example 2:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 0; % classic LBP
% options.samples = 8; % number of neighbor samples
% options.mappingtype = 'ri'; % rotation-invariant LBP
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% figure(1);imshow(J,[]) % image to be analyzed
% [X,Xn] = Bfx_lbp(J,[],options); % LBP features
% bar(X) % histogram
%
% Example 3:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 4:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 16; % number of neighbor samples
% options.sk = 0.5; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % semantic LBP features
% bar(X) % histogram
%
% Example 5:
% options.vdiv = 1; % one vertical divition
% options.hdiv = 1; % one horizontal divition
% options.semantic = 1; % semantic LBP
% options.samples = 8; % number of neighbor samples
% options.sk = 0.25; % angle sampling
% options.weight = 9; % angle sampling
% I = imread('testimg1.jpg'); % input image
% J = I(120:219,120:239,2); % region of interest (green)
% [X,Xn] = Bfx_lbp(J,[],options); % weighted LBP features
% bar(X) % histogram
% See also Bfx_gabor, Bfx_clp, Bfx_fourier, Bfx_dct.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
%
function [X,Xn,options] = Bfx_lbp(I,R,options)
if nargin==2;
options = R;
R = ones(size(I));
end
vdiv = options.vdiv;
hdiv = options.hdiv;
if ~isfield(options,'show')
options.show = 0;
end
if ~isfield(options,'normalize')
options.normalize = 0;
end
if options.show == 1
disp('--- extracting local binary patterns features...');
end
if ~isfield(options,'samples')
options.samples = 8;
end
if ~isfield(options,'integral')
options.integral = 0;
end
if ~isfield(options,'radius')
options.radius = log(options.samples)/log(2)-1;
end
if ~isfield(options,'semantic')
options.semantic = 0;
end
if ~isfield(options,'weight')
options.weight = 0;
end
LBPst = 'LBP';
if options.semantic>0
if ~isfield(options,'sk')
options.sk = 1;
end
mapping = getsmapping(options.samples,options.sk);
LBPst = ['s' LBPst];
st='8x8';
else
% mapping = getmapping(8,'u2');
if ~isfield(options,'mappingtype')
options.mappingtype = 'u2';
end
st = sprintf('%d,%s',options.samples,options.mappingtype);
mapping = getmapping(options.samples,options.mappingtype);
end
% get lbp image
if ~isempty(R);
I(R==0) = 0;
end
code_img = lbp(I,options.radius,options.samples,mapping,'');
[n1,n2] = size(code_img);
[N,M] = size(I);
Ilbp = zeros(size(I));
i1 = round((N-n1)/2);
j1 = round((M-n2)/2);
Ilbp(i1+1:i1+n1,j1+1:j1+n2) = code_img;
options.Ilbp = Ilbp;
if options.integral == 1
options.Hx = Bim_inthist(Ilbp+1,options.maxD);
end
ylen = round(n1/vdiv);
xlen = round(n2/hdiv);
% split image into blocks (saved as columns)
grid_img = im2col(code_img,[ylen, xlen], 'distinct');
if options.weight>0
LBPst = ['w' LBPst];
mt = 2*options.radius-1;
mt2 = mt^2;
Id = double(I);
switch options.weight
case 1
W = abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id);
case 2
W = (abs(conv2(Id,ones(mt,mt)/mt2,'same')-Id))./(Id+1);
case 3
W = abs(medfilt2(Id,[mt mt])-Id);
case 4
W = abs(medfilt2(Id,[mt mt])-Id)./(Id+1);
case 5
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id);
case 6
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
case 7
Id = conv2(Id,ones(mt,mt)/mt2,'same');
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
case 8
Id = medfilt2(Id,[mt mt]);
W = abs(ordfilt2(Id,mt2,ones(mt,mt))-Id)./(Id+1);
case 9
Id = medfilt2(Id,[mt mt]);
W = abs(ordfilt2(Id,mt2-1,ones(mt,mt))-Id)./(Id+1);
otherwise
error('Bfx_lbp does not recognice options.weight = %d.',options.weight);
end
W = W(mt+1:end-mt,mt+1:end-mt);
grid_W = im2col(W,[ylen, xlen], 'distinct');
nwi = mapping.num;
nwj = size(grid_W,2);
nwk = size(grid_W,1);
desc = zeros(nwi,nwj);
for j=1:nwj
x = grid_img(:,j)+1;
y = grid_W(:,j);
d = zeros(nwi,1);
for k=1:nwk
d(x(k))=d(x(k))+y(k);
% d(x(k))=d(x(k))+1; % normal LBP each LBP has equal weight
end
desc(:,j) = d;
end
else
desc = hist(double(grid_img), 0:mapping.num-1);
% calculate coordinates of descriptors as if I was square w/ side=1
end
dx = 1.0/hdiv;
dy = 1.0/vdiv;
x = dx/2.0: dx :1.0-dx/2.0;
y = dy/2.0: dy :1.0-dy/2.0;
options.x = x;
options.y = y;
if hdiv*vdiv>1
D = desc';
else
D = desc;
end
[M,N] = size(D);
Xn = char(zeros(N*M,24));
X = zeros(1,N*M);
k=0;
for i=1:M
for j=1:N
k = k+1;
s = sprintf('%s(%d,%d)[%s] ',LBPst,i,j,st);
Xn(k,:) = s(1:24);
X(k) = D(i,j);
end
end
if options.normalize
X = X/sum(X);
end
end
%GETMAPPING returns a structure containing a mapping table for LBP codes.
% MAPPING = GETMAPPING(SAMPLES,MAPPINGTYPE) returns a
% structure containing a mapping table for
% LBP codes in a neighbourhood of SAMPLES sampling
% points. Possible values for MAPPINGTYPE are
% 'u2' for uniform LBP
% 'ri' for rotation-invariant LBP
% 'riu2' for uniform rotation-invariant LBP.
%
% Example:
% I=imread('rice.tif');
% MAPPING=getmapping(16,'riu2');
% LBPHIST=lbp(I,2,16,MAPPING,'hist');
% Now LBPHIST contains a rotation-invariant uniform LBP
% histogram in a (16,2) neighbourhood.
%
function mapping = getmapping(samples,mappingtype)
% Version 0.1.1
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% 0.1.1 Changed output to be a structure
% Fixed a bug causing out of memory errors when generating rotation
% invariant mappings with high number of sampling points.
% Lauge Sorensen is acknowledged for spotting this problem.
table = 0:2^samples-1;
newMax = 0; %number of patterns in the resulting LBP code
index = 0;
if strcmp(mappingtype,'u2') %Uniform 2
newMax = samples*(samples-1) + 3;
for i = 0:2^samples-1
j = bitset(bitshift(i,1,samples),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples)); %number of 1->0 and
%0->1 transitions
%in binary string
%x is equal to the
%number of 1-bits in
%XOR(x,Rotate left(x))
if numt <= 2
table(i+1) = index;
index = index + 1;
else
table(i+1) = newMax - 1;
end
end
end
if strcmp(mappingtype,'ri') %Rotation invariant
tmpMap = zeros(2^samples,1) - 1;
for i = 0:2^samples-1
rm = i;
r = i;
for j = 1:samples-1
r = bitset(bitshift(r,1,samples),1,bitget(r,samples)); %rotate
%left
if r < rm
rm = r;
end
end
if tmpMap(rm+1) < 0
tmpMap(rm+1) = newMax;
newMax = newMax + 1;
end
table(i+1) = tmpMap(rm+1);
end
end
if strcmp(mappingtype,'riu2') %Uniform & Rotation invariant
newMax = samples + 2;
for i = 0:2^samples - 1
j = bitset(bitshift(i,1,samples),1,bitget(i,samples)); %rotate left
numt = sum(bitget(bitxor(i,j),1:samples));
if numt <= 2
table(i+1) = sum(bitget(i,1:samples));
else
table(i+1) = samples+1;
end
end
end
mapping.table=table;
mapping.samples=samples;
mapping.num=newMax;
end
% LBP returns the local binary pattern image or LBP histogram of an image.
% J = LBP(I,R,N,MAPPING,MODE) returns either a local binary pattern
% coded image or the local binary pattern histogram of an intensity
% image I. The LBP codes are computed using N sampling points on a
% circle of radius R and using mapping table defined by MAPPING.
% See the getmapping function for different mappings and use 0 for
% no mapping. Possible values for MODE are
% 'h' or 'hist' to get a histogram of LBP codes
% 'nh' to get a normalized histogram
% Otherwise an LBP code image is returned.
%
% J = LBP(I) returns the original (basic) LBP histogram of image I
%
% J = LBP(I,SP,MAPPING,MODE) computes the LBP codes using n sampling
% points defined in (n * 2) matrix SP. The sampling points should be
% defined around the origin (coordinates (0,0)).
%
% Examples
% --------
% I=imread('rice.png');
% mapping=getmapping(8,'u2');
% H1=LBP(I,1,8,mapping,'h'); %LBP histogram in (8,1) neighborhood
% %using uniform patterns
% subplot(2,1,1),stem(H1);
%
% H2=LBP(I);
% subplot(2,1,2),stem(H2);
%
% SP=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
% I2=LBP(I,SP,0,'i'); %LBP code image using sampling points in SP
% %and no mapping. Now H2 is equal to histogram
% %of I2.
function result = lbp(varargin) % image,radius,neighbors,mapping,mode)
% Version 0.3.2
% Authors: Marko Heikkila and Timo Ahonen
% Changelog
% Version 0.3.2: A bug fix to enable using mappings together with a
% predefined spoints array
% Version 0.3.1: Changed MAPPING input to be a struct containing the mapping
% table and the number of bins to make the function run faster with high number
% of sampling points. Lauge Sorensen is acknowledged for spotting this problem.
% Check number of input arguments.
error(nargchk(1,5,nargin));
image=varargin{1};
d_image=double(image);
if nargin==1
spoints=[-1 -1; -1 0; -1 1; 0 -1; -0 1; 1 -1; 1 0; 1 1];
neighbors=8;
mapping=0;
mode='h';
end
if (nargin == 2) && (length(varargin{2}) == 1)
error('Input arguments');
end
if (nargin > 2) && (length(varargin{2}) == 1)
radius=varargin{2};
neighbors=varargin{3};
spoints=zeros(neighbors,2);
% Angle step.
a = 2*pi/neighbors;
for i = 1:neighbors
spoints(i,1) = -radius*sin((i-1)*a);
spoints(i,2) = radius*cos((i-1)*a);
end
if(nargin >= 4)
mapping=varargin{4};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 5)
mode=varargin{5};
else
mode='h';
end
end
if (nargin > 1) && (length(varargin{2}) > 1)
spoints=varargin{2};
neighbors=size(spoints,1);
if(nargin >= 3)
mapping=varargin{3};
if(isstruct(mapping) && mapping.samples ~= neighbors)
error('Incompatible mapping');
end
else
mapping=0;
end
if(nargin >= 4)
mode=varargin{4};
else
mode='h';
end
end
% Determine the dimensions of the input image.
[ysize xsize] = size(image);
miny=min(spoints(:,1));
maxy=max(spoints(:,1));
minx=min(spoints(:,2));
maxx=max(spoints(:,2));
% Block size, each LBP code is computed within a block of size bsizey*bsizex
bsizey=ceil(max(maxy,0))-floor(min(miny,0))+1;
bsizex=ceil(max(maxx,0))-floor(min(minx,0))+1;
% Coordinates of origin (0,0) in the block
origy=1-floor(min(miny,0));
origx=1-floor(min(minx,0));
% Minimum allowed size for the input image depends
% on the radius of the used LBP operator.
if(xsize < bsizex || ysize < bsizey)
error('Too small input image. Should be at least (2*radius+1) x (2*radius+1)');
end
% Calculate dx and dy;
dx = xsize - bsizex;
dy = ysize - bsizey;
% Fill the center pixel matrix C.
C = image(origy:origy+dy,origx:origx+dx);
d_C = double(C);
bins = 2^neighbors;
% Initialize the result matrix with zeros.
result=zeros(dy+1,dx+1);
%Compute the LBP code image
for i = 1:neighbors
y = spoints(i,1)+origy;
x = spoints(i,2)+origx;
% Calculate floors, ceils and rounds for the x and y.
fy = floor(y); cy = ceil(y); ry = round(y);
fx = floor(x); cx = ceil(x); rx = round(x);
% Check if interpolation is needed.
if (abs(x - rx) < 1e-6) && (abs(y - ry) < 1e-6)
% Interpolation is not needed, use original datatypes
N = image(ry:ry+dy,rx:rx+dx);
D = N >= C;
else
% Interpolation needed, use double type images
ty = y - fy;
tx = x - fx;
% Calculate the interpolation weights.
w1 = (1 - tx) * (1 - ty);
w2 = tx * (1 - ty);
w3 = (1 - tx) * ty ;
w4 = tx * ty ;
% Compute interpolated pixel values
N = w1*d_image(fy:fy+dy,fx:fx+dx) + w2*d_image(fy:fy+dy,cx:cx+dx) + ...
w3*d_image(cy:cy+dy,fx:fx+dx) + w4*d_image(cy:cy+dy,cx:cx+dx);
D = N >= d_C;
end
% Update the result matrix.
v = 2^(i-1);
result = result + v*D;
end
%Apply mapping if it is defined
if isstruct(mapping)
bins = mapping.num;
for i = 1:size(result,1)
for j = 1:size(result,2)
result(i,j) = mapping.table(result(i,j)+1);
end
end
end
if (strcmp(mode,'h') || strcmp(mode,'hist') || strcmp(mode,'nh'))
% Return with LBP histogram if mode equals 'hist'.
result=hist(result(:),0:(bins-1));
if (strcmp(mode,'nh'))
result=result/sum(result);
end
else
%Otherwise return a matrix of unsigned integers
if ((bins-1)<=intmax('uint8'))
result=uint8(result);
elseif ((bins-1)<=intmax('uint16'))
result=uint16(result);
else
result=uint32(result);
end
end
end
% Mapping for sLBB
function mapping = getsmapping(N,sk)
M = 2^N;
samples = N;
len = zeros(M,1);
ang = zeros(M,1);
for x=0:(M-1)
k = x+1;
j = bitset(bitshift(x,1,samples),1,bitget(x,samples));
numt = sum(bitget(bitxor(x,j),1:samples));
c = numt;
if c>2
len(k)=-1;
ang(k)=-1;
else
s = bitget(x,1:samples);
len(k) = sum(s);
if c==0
ang(k)=0;
else
r = 0;
while (s(1)~=0) || (s(N)~=1)
s = [s(2:N) s(1)];
r = r+1;
end
ii = find(s==1);
a = mean(ii)+r;
if a>N
a=a-N;
end
ang(k) = round(sk*a)-1;
end
end
% fprintf('%4d: %s (%d,%d)\n',x,dec2bin(x,N),len(k),ang(k)); pause
end
Ma = max(ang)+1;
map = len*Ma+ang-Ma+1;
n = max(map)+1;
map(ang==-1) = n;
map(1) = 0;
mapping.table = map';
mapping.samples = N;
mapping.num = n+1;
end
|
github
|
domingomery/Balu-master
|
Bfx_geo.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_geo.m
| 2,922 |
utf_8
|
446405d4d737c6637957ae9496e02e5a
|
% [X,Xn] = Bfx_geo(R,options)
% [X,Xn] = Bfx_geo(L,options)
%
% Toolbox: Balu
%
% Gemteric feature extraction.
%
% This function calls gemetric feature extraction procedures of binary
% image R or labelled image L.
%
% X is the feature matrix (one feature per column, one sample per row),
% Xn is the list with the names of these features (see Example
% to see how it works).
%
% Example 1: Extraction of one region image
% b(1).name = 'hugeo'; b(1).options.show=1; % Hu moments
% b(2).name = 'flusser'; b(2).options.show=1; % Flusser moments
% b(3).name = 'fourierdes'; b(3).options.show=1; % Fourier
% b(3).options.Nfourierdes=12; % descriptors
% options.b = b;
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_geo(R,options); % geometric features
% Bio_printfeatures(X,Xn)
%
% Example 2: Extraction of multiple regions image
% b(1).name = 'hugeo'; b(1).options.show=1; % Hu moments
% b(2).name = 'basicgeo'; b(2).options.show=1; % basic geometric fetaures
% b(3).name = 'fourierdes'; b(3).options.show=1; % Fourier
% b(3).options.Nfourierdes=12; % descriptors
% options.b = b;
% I = imread('rice.png'); % input image
% [R,m] = Bim_segmowgli(I,ones(size(I)),40,1.5); % segmentation
% [X,Xn] = Bfx_geo(R,options); % geometric features
% figure; hist(X(:,12));xlabel([Xn(12,:)]) % area histogramm
% ii = find(abs(X(:,20))<15); % rice orientation
% K = zeros(size(R)); % between -15 and 15 grad
% for i=1:length(ii);K=or(K,R==ii(i));end
% figure; imshow(K);title('abs(orientation)<15 grad')
%
% See also Bfx_basicgeo, Bfx_hugeo, Bfx_flusser, Bfx_gupta,
% Bfx_fitellipse, Bfx_fourierdes, Bfx_files.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_geo(R,options)
if isempty(R)
error('Bfx_geo: R is empty. Geometric features without segmentation has no sense.');
else
b = options.b;
n = length(b);
m = int16(max(R(:)));
X = [];
for j=1:m
Rj = R==j;
Xj = [];
Xnj = [];
for i=1:n
s = b(i).name;
if sum(s(1:3)=='Bfx')~=3
s = ['Bfx_' s];
end
if ~exist(s,'file')
error(sprintf('Bfx_geo: function %s does not exist.',b(i).name))
end
[Xi,Xni] = feval(s,Rj,b(i).options);
Xj = [Xj Xi];
Xnj = [Xnj; Xni];
end
X = [X;Xj];
end
Xn = Xnj;
end
|
github
|
domingomery/Balu-master
|
Bfx_int.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_int.m
| 5,881 |
utf_8
|
580daf50396b7da9cbd6894e5fc8da80
|
% [X,Xn] = Bfx_int(I,R,b)
%
% Toolbox: Balu
%
% Intensity feature extraction.
%
% This function calls intensity feature extraction procedures of image I
% according binary image R. See example to see how it works.
%
% X is the feature matrix (one feature per column, one sample per row),
% Xn is the list of feature names (see Examples to see how it works).
%
% Example 1: Extraction of one region image in grayvalue images
% b(1).name = 'gabor'; b(1).options.show=1; % Gabor features
% b(1).options.Lgabor = 8; % number of rotations
% b(1).options.Sgabor = 8; % number of dilations (scale)
% b(1).options.fhgabor = 2; % highest frequency of interest
% b(1).options.flgabor = 0.1; % lowest frequency of interest
% b(1).options.Mgabor = 21; % mask size
% b(2).name = 'basicint'; b(2).options.show=1; % Basic intensity features
% b(3).name = 'lbp'; b(3).options.show=1; % LBP
% b(3).options.vdiv = 2; % vertical div
% b(3).options.hdiv = 2; % horizontal div
% options.b = b;
% options.colstr = 'i'; % gray image
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% J = rgb2gray(I); % grayvalue image
% [X,Xn] = Bfx_int(J,R,options); % intensity features
% Bio_printfeatures(X,Xn)
%
% Example 2: Extraction of multiple regions image
% b(1).name = 'huint'; b(1).options.show=1; % Hu moments
% b(2).name = 'basicint'; b(2).options.show=1; % Basic intensity features
% b(3).name = 'lbp'; b(3).options.show=1; % LBP
% b(3).options.vdiv = 2; % vertical div
% b(3).options.hdiv = 2; % horizontal div
% b(4).name = 'contrast'; b(4).options.show = 1; % Contrast
% b(4).options.neighbor = 1; % neighborhood is a window
% b(4).options.param = 1.5; % 1.5 height x 1.5 width
% I = imread('rice.png'); % input image
% [R,m] = Bim_segmowgli(I,ones(size(I)),40,1.5); % segmentation
% options.b = b;
% options.colstr = 'i'; % gray image
% [X,Xn] = Bfx_int(I,R,options); % intensity features
% figure; hist(X(:,9));xlabel([Xn(9,:)]) % std dev histogramm
% figure; hist(X(:,253));xlabel([Xn(253,:)]) % contrast Ks histogramm
%
% Example 3: Extraction of one region image in RGB images
% b(1).name = 'gabor'; b(1).options.show=1; % Gabor features
% b(1).options.Lgabor = 8; % number of rotations
% b(1).options.Sgabor = 8; % number of dilations (scale)
% b(1).options.fhgabor = 2; % highest frequency of interest
% b(1).options.flgabor = 0.1; % lowest frequency of interest
% b(1).options.Mgabor = 21; % mask size
% b(2).name = 'basicint'; b(2).options.show=1; % Basic intensity features
% b(3).name = 'lbp'; b(3).options.show=1; % LBP
% b(3).options.vdiv = 2; % vertical div
% b(3).options.hdiv = 2; % horizontal div
% options.b = b;
% options.colstr = 'rgb'; % R image
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_int(I,R,options); % intensity features
% Bio_printfeatures(X,Xn)
%
% See also Bfx_basicint, Bfx_haralick, Bfx_gabor, Bfx_dct, Bfx_fourier,
% Bfx_huint, Bfx_lbp, Bfx_contrast, Bfx_clp, Bfx_phog,
% Bfx_files.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_int(I,R,options)
I = double(I);
[N,M,P] = size(I);
c = size(I,3);
if nargin==2;
options = R;
R = ones(size(I));
end
if isempty(R)
R = ones(N,M);
end
b = options.b;
colstr = options.colstr;
n = length(b);
m = int16(max(R(:)));
X = [];
Xn = [];
for j=1:m
Rj = R==j;
Xj = [];
if j==1
Xn = [];
end
for i=1:n
s = b(i).name;
if sum(s(1:2)=='Bf')~=2
s = ['Bfx_' s];
end
if ~exist(s,'file')
error(sprintf('Bfx_extraction: function %s does not exist.',b(i).name))
end
% if or(compare(s,'Bfx_lbphogi')==0,compare(s,'Bfx_hogi')==0)
% if (compare(s(1:8),'Bfx_lbpi')*compare(s,'Bfx_lbphogi')*compare(s,'Bfx_hogi'))==0
if (compare(s,'Bfx_lbpi')*compare(s,'Bfx_lbphogi')*compare(s,'Bfx_hogi'))==0
c = 1;
ifull = 1;
else
c = P;
ifull = 0;
end
for k=1:c
% for i=1:n
if ifull
% tic
[Xk,Xnk] = feval(s,I,Rj,b(i).options);
% toc
else
[Xk,Xnk] = feval(s,I(:,:,k),Rj,b(i).options);
end
Xj = [Xj Xk];
if j==1
nk = length(Xk);
Xns = [ones(nk,1)*[colstr(k) '-'] Xnk ];
Xn = [Xn; Xns(:,1:24)];
end
end
end
X = [X;Xj];
end
|
github
|
domingomery/Balu-master
|
Bfx_all.m
|
.m
|
Balu-master/FeatureExtraction/Bfx_all.m
| 499 |
utf_8
|
dd2f13d940eb76ecc4bb1cbf3cea7857
|
% [X,Xn] = Bfx_all(I,R,options)
% [X,Xn] = Bfx_all(I,options)
%
% Toolbox: Balu
% All pixels.
%
% X is the features vector, Xn is the list feature names (see Example to
% see how it works).
%
%
% Example:
% I = imread('testimg1.jpg'); % input image
% [X,Xn] = Bfx_all(I); % all pixels
%
%
% (c) D.Mery, PUC-DCC, 2012
% http://dmery.ing.puc.cl
function [X,Xn] = Bfx_all(I,R,options)
X = I(:)';
Xn = zeros(length(X),24);
|
github
|
domingomery/Balu-master
|
num2fixstr.m
|
.m
|
Balu-master/Miscellaneous/num2fixstr.m
| 347 |
utf_8
|
842ba059f8799b2c4a46e890de6cee74
|
%function st = num2fixstr(i,d)
%
% This function converts an integer i in a string
% st with a fixed number of characters (with '0'
% filled from left.
%
% Example:
% st = num2fixstr(3,4) % returns '0003'
%
% D.Mery, Aug-2012
% http://dmery.ing.puc.cl
%
function st = num2fixstr(i,d)
s = [char('0'*ones(1,d)) num2str(i)];
st = s(end-d+1:end);
|
github
|
domingomery/Balu-master
|
enterpause.m
|
.m
|
Balu-master/Miscellaneous/enterpause.m
| 311 |
utf_8
|
a24fcf539906b8dc95c0e17d240bd116
|
% Toolbox: Balu
% enterpause display "press <Enter> to continue..." and wait for <Enter>
% enterpause(t) means pause(t)
%
% D.Mery, PUC-DCC, Apr. 2008
% http://dmery.ing.puc.cl
function enterpause(t)
if ~exist('t','var')
disp('press <Enter> to continue...')
pause
else
pause(t)
end
|
github
|
domingomery/Balu-master
|
posrandom.m
|
.m
|
Balu-master/Miscellaneous/posrandom.m
| 279 |
utf_8
|
e209ba3e41ac82dd560a8ff602aad8af
|
% It changes the order of the columns or rows of x randomly. Variable dim
% selects the dimension along which to change.
function [x_new,j] = posrandom(x,dim)
nx = size(x,dim);
r = rand(nx,1);
[i,j] = sort(r);
if dim==1
x_new = x(j,:);
else
x_new = x(:,j);
end
|
github
|
domingomery/Balu-master
|
cart2polpos.m
|
.m
|
Balu-master/Miscellaneous/cart2polpos.m
| 870 |
utf_8
|
b291b2687cca1f26635f22687f1ebc7b
|
% [t,r] = cart2polpos(x,y,s)
%
% Toolbox: Balu
% transform Cartesian to polar coordinates. The angle t will be always
% positive. If s==0, the t is between 0 and 2*pi. If s > 0, the angles are
% subdivided into s bins, and t means in which bin the angle is located.
%
% Examples:
% [t,r] = cart2polpos(1,1) % it is the same as [t,r] = cart2polpos(1,1,0)
% the result for t is pi/4 and for r = sqrt(2)/2
%
% [t,r] = cart2polpos(1,-1) % it is the same as [t,r] = cart2polpos(1,1,0)
% the result for t is 7*pi/4 and for r = sqrt(2)/2
%
% [t,r] = cart2polpos(1,-1,4) % example that computes the quadrant
% the result for t is 4 and for r = sqrt(2)/2
%
% D.Mery, PUC-DCC, 2014
% http://dmery.ing.puc.cl
function [t,r] = cart2polpos(x,y,s)
if ~exist('s','var')
s=0;
end
[t,r] = cart2pol(x,y);
t(t<0) = t(t<0)+2*pi;
if s>0
t = fix(t/2/pi*s)+1;
end
|
github
|
domingomery/Balu-master
|
distxy.m
|
.m
|
Balu-master/Miscellaneous/distxy.m
| 2,111 |
utf_8
|
549e0d02ce5a7c63b7150ee7d5e67ab1
|
% function D = distxy(x,y,method)
%
% Toolbox: Balu
%
% It computes distances of each row of x with each row of y
% x is a Nx x n matrix, y is a Ny x n matrix
% D is a Nx x Ny matrix, where D(i,j) = norm(x(i,:)-y(j,:));
% method = 1: it computes row per row
% 2: it computes each row of x with all rows of y
% 3: it computes all rows of x with all rows of y
% method 3 is faster than method 2, and this faster than method 1
% method 3 requires more memory than method 2, and this more than method 1
% method 3 is the default.
%
% Example:
% x = rand(100,7);
% y = rand(100,7);
% D = distxy(x,y); % default is method 3
%
% D.Mery, PUC-DCC, 2012
% http://dmery.ing.puc.cl
function D = distxy(x,y,method)
if ~exist('method','var')
method = 3;
end
[Nx,nx] = size(x);
[Ny,ny] = size(y);
if nx~=ny
error('number of elements of columns of x and y must be same...')
end
n = nx;
D = zeros(Nx,Ny);
switch method
case 1
% ff = Bio_statusbar('distances');
for i=1:Nx
% Bio_statusbar(i/Nx,ff);
for j=1:Ny
D(i,j) = norm(x(i,:)-y(j,:));
end
end
% delete(ff)
case 2
% ff = Bio_statusbar('distances');
for i=1:Nx
% Bio_statusbar(i/Nx,ff);
di = ones(Ny,1)*x(i,:)-y;
D(i,:) = sqrt(sum(di.*di,2)');
end
% delete(ff)
case 3
xs = zeros(1,n,Nx);
xs(:) = x';
Sx = repmat(xs,[Ny 1 1]);
Sx = Sx-repmat(y,[1 1 Nx]);
Sx = Sx.*Sx;
R = sqrt(sum(Sx,2));
Dr = zeros(Ny,Nx);
Dr(:) = R;
D = Dr';
case 4
kd = vl_kdtreebuild(x');
[i,d] = vl_kdtreequery(kd,x',y','NumNeighbors',Nx);
[j,k] = sort(i);
D = sqrt(d(k));
case 5 % it is like 3 but with out sqrt
xs = zeros(1,n,Nx);
xs(:) = x';
Sx = repmat(xs,[Ny 1 1]);
Sx = Sx-repmat(y,[1 1 Nx]);
Sx = Sx.*Sx;
R = sum(Sx,2);
Dr = zeros(Ny,Nx);
Dr(:) = R;
D = Dr';
end
|
github
|
domingomery/Balu-master
|
andsift.m
|
.m
|
Balu-master/Miscellaneous/andsift.m
| 714 |
utf_8
|
783f127c786620dfefdb1c28025e7cb7
|
% It separates sift descriptors into descriptors that belong to the region
% of interest R. R is a binary image. fi,di are the frames and descriptors
% of pixels = i (for i=0,1). (ff,dd) is the transposed output of vl_sift.
function [f1,d1,f0,d0,i1,i0] = andsift(ff,dd,R)
f = ff';
d = dd';
[N,M] = size(R);
ii = round(f(2,:));
jj = round(f(1,:));
kk = (ii<1)|(jj<1)|(ii>N)|(jj>M);
ii(kk) = [];
jj(kk) = [];
kk = sub2ind([N M],ii,jj);
if ~isempty(kk)
r1 = R(kk);
i1 = find(r1==1);
i0 = find(r1==0);
f1 = f(:,i1);
d1 = d(:,i1);
f0 = f(:,i0);
d0 = d(:,i0);
else
f1 = [];
d1 = [];
f0 = f;
d0 = d;
end
f1 = f1';
d1 = d1';
f0 = f0';
d0 = d0';
|
github
|
domingomery/Balu-master
|
howis.m
|
.m
|
Balu-master/Miscellaneous/howis.m
| 489 |
utf_8
|
08b5567797d8a802206bf707280552b5
|
% howis(x)
%
% Toolbox: Balu
% How is x? it displays min, max, size, class of x.
% If x is a structure, it display the field names.
%
% D.Mery, PUC-DCC, Jul 2009-2012
% http://dmery.ing.puc.cl
%
function howis(x)
if isstruct(x)
disp('Structure:')
x
else
fprintf('%s\n',['min = ' num2str(min(x(:)))])
fprintf('%s\n',['max = ' num2str(max(x(:)))])
fprintf('%s\n',['size = ' num2str(size(x))])
fprintf('%s\n',['class = ' class(x)])
end
|
github
|
domingomery/Balu-master
|
Bfx_centroid.m
|
.m
|
Balu-master/Examples/Bfx_centroid.m
| 1,071 |
utf_8
|
535fc870f63dc5c0db9c6cd5bd5a99fc
|
% [X,Xn] = Bfx_centroid(R,options)
%
% Toolbox: Balu
%
% Centroid of a region.
%
% options.show = 1 display mesagges.
%
% X(1) is centroid-i, X(2) is centroid-j
% Xn is the list of the n feature names.
%
% Example (Centroid of a region)
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% imshow(R);
% options.show = 1;
% [X,Xn] = Bfx_centroid(R,options);
% Bio_printfeatures(X,Xn)
%
% See also Bfx_basicgeo, Bfx_gupta, Bfx_fitellipse, Bfx_flusser,
% Bfx_hugeo.
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function [X,Xn] = Bfx_centroid(R,options)
if options.show == 1
disp('--- extracting centroid...');
end
[Ireg,Jreg] = find(R==1); % pixels in the region
ic = mean(Ireg);
jc = mean(Jreg);
X = [ic jc];
Xn = [ 'Centroid i '
'Centroid j ']; % 24 characters per name
if options.show == 1
clf
imshow(R)
hold on
plot(X(2),X(1),'rx')
enterpause
end
|
github
|
domingomery/Balu-master
|
Bex_sfssvm.m
|
.m
|
Balu-master/Examples/Bex_sfssvm.m
| 1,028 |
utf_8
|
e44139735f36fbd1b9b4572ee25ad370
|
% s = Bex_sfssvm(f,d,m)
%
% Toolbox: Balu
% Example: Feature selection using Balu algorithm based on SFS and SVM.
%
% Bfs_balu has three steps:
% (1) normalizes (using Bft_norm),
% (2) cleans (using Bfs_clean), and
% (3) selects features (using Bfs_sfs).
%
% Example:
% load datareal
% s = Bex_sfssvm(f,d,6)
% fn(s,:)
%
% See also Bfs_balu
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function s = Bex_sfssvm(f,d,m)
op.m = m; % m features will be selected
op.s = 0.75; % only 75% of sample will be used
op.show = 1; % display results
op.b.name = 'svm'; % SFS with SVM
op.b.options.kernel = 4; % SVM-RBF
s = Bfs_balu(f,d,op); % index of selected features
X = f(:,s); % selected features
op = op.b.options;
[X1,d1,X2,d2] = Bds_stratify(X,d,0.75);
d2s = Bcl_svm(X1,d1,X2,op);
p = Bev_performance(d2,d2s); % performance with lsef
fprintf('Performance with SFS and SVM = %5.4f\n',p)
|
github
|
domingomery/Balu-master
|
Bex_decisionline.m
|
.m
|
Balu-master/Examples/Bex_decisionline.m
| 707 |
utf_8
|
49c2a2387a4afed7977a6df45a94a357
|
% Bex_decisionline(X,d,bcl)
%
% Toolbox: Balu
% Example: Decision lines of features X with labels d for classifiers
% defined in bcl
%
% Example:
% load datagauss % simulated data (2 classes, 2 features)
% Xn = ['x_1';'x_2'];
% bcl(1).name = 'knn'; bcl(1).options.k = 5; % KNN with 5 neighbors
% bcl(2).name = 'lda'; bcl(2).options.p = []; % LDA
% bcl(3).name = 'svm'; bcl(3).options.kernel = 3; % rbf-SVM
% Bex_decisionline(X,d,bcl)
%
% See also Bio_decisionline
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function Bex_decisionline(X,d,bcl)
op = Bcl_structure(X,d,bcl);
Bio_decisionline(X,d,['x1';'x2'],op);
|
github
|
domingomery/Balu-master
|
Bex_exsknn.m
|
.m
|
Balu-master/Examples/Bex_exsknn.m
| 1,344 |
utf_8
|
1cc04f149ac136e42e9984448caeee04
|
% s = Bex_exsknn(f,d,m)
%
% Toolbox: Balu
% Example: Feature selection using exhaustive search and KNN.
%
% Bfs_balu has three steps:
% (1) normalizes (using Bft_norm),
% (2) cleans (using Bfs_clean), and
% (3) selects features (using Bfs_sfs).
%
% Example:
% load datareal
% s = Bex_exsknn(f,d)
% fn(s,:)
%
% See also Bfs_balu
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function s = Bex_exsknn(f,d)
op.m = 10; % 10 features will be pre-selected
op.s = 0.75; % only 75% of sample will be used
op.show = 1; % display results
op.b.name = 'fisher'; % SFS with fisher
s1 = Bfs_balu(f,d,op); % index of selected features
X1 = f(:,s1); % selected features
op.m = 4; % 3 features will be selected
op.show = 1; % display results
op.b.name = 'knn'; % SFS with KNN
op.b.options.k = 5; % 5 neighbors
s2 = Bfs_exsearch(X1,d,op); % index of selected features
s = s1(s2); % list of feature names
X = f(:,s);
op = op.b.options;
[X1,d1,X2,d2] = Bds_stratify(X,d,0.75);
d2s = Bcl_knn(X1,d1,X2,op);
p = Bev_performance(d2,d2s); % performance with lsef
fprintf('Performance with exhaustive search and KNN = %5.4f\n',p)
|
github
|
domingomery/Balu-master
|
Bex_fscombination.m
|
.m
|
Balu-master/Examples/Bex_fscombination.m
| 3,104 |
utf_8
|
4470da31ed3e0aa327ccb8d95e0cbbff
|
% Bex_fscombination
%
% Toolbox: Balu
% Combination of feature selection algorithms.
%
% This example shows how to combine different feature selection algorthm
% in orde to obtain the highest performance.
%
% The evaluation of the performance is using a simple LDA classifier
%
% Example:
% load datareal
% Bex_fscombination
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function Bex_fscombination(f,d)
op_sfs.m = 40; % 40 features will be selected
op_sfs.s = 1; % 100% of sample will be used
op_sfs.show = 1; % display results
op_sfs.b.name = 'fisher'; % SFS with Fisher
s = Bfs_balu(f,d,op_sfs); % index of selected features
XSFS40 = f(:,s); % SFS first 40 features
XPCA40 = Bft_pca(f,40); % first 40 principal components
disp('1. Feature selection with SFS...')
X = XSFS40(:,1:6);
lda_performance(X,d) % function at the end of this code
disp('2. Feature selection with PCA only...')
X = XPCA40(:,1:6);
lda_performance(X,d)
disp('3. Feature selection with PCA of SFS...')
X = Bft_pca(XSFS40,6); % first 6 principal components
lda_performance(X,d) % function at the end of this code
disp('4. Feature selection with SFS of PCA and SFS...')
X1 = [XPCA40 XSFS40];
op_sfs.m = 6; % 6 features will be selected
s = Bfs_balu(X1,d,op_sfs); % index of selected features
X = X1(:,s);
lda_performance(X,d) % function at the end of this code
disp('5. Feature selection with SFS of PCA of SFS...')
X1 = Bft_pca(XSFS40,20); % X1 from last step
X2 = [X1 XSFS40];
s = Bfs_balu(X2,d,op_sfs); % index of selected features
X = X2(:,s);
lda_performance(X,d) % function at the end of this code
disp('6. Feature selection with SFS of PCA and all features...')
% (c) Esteban Cortazar, 2010
X1 = [XPCA40 f];
s = Bfs_balu(X1,d,op_sfs); % index of selected features
X = X1(:,s);
lda_performance(X,d) % function at the end of this code
disp('7. Feature selection with LSEF of SFS...')
op2.m = 6; % 5 features will be selected
op2.show = 0; % display results
s = Bfs_lsef(XSFS40,op2); % index of selected features
X = X1(:,s);
lda_performance(X,d) % function at the end of this code
disp('8. Feature selection with FOSMOD of SFS...')
s = Bfs_fosmod(XSFS40,op2);% data from last step
X = X1(:,s);
lda_performance(X,d) % function at the end of this code
disp('9. Feature selection with PLSR of SFS...')
X = Bft_plsr(XSFS40,d,6);
lda_performance(X,d) % function at the end of this code
end
function lda_performance(X,d)
op_lda.p = [];
ds = Bcl_lda(X,d,X,op_lda);
Bev_performance(d,ds)
enterpause
end
|
github
|
domingomery/Balu-master
|
Bex_lsef.m
|
.m
|
Balu-master/Examples/Bex_lsef.m
| 1,379 |
utf_8
|
80041176a9d5dd6cf64549a184d4ad9b
|
% s = Bex_lsef(f,d,m)
%
% Toolbox: Balu
% Example: Feature selection using lsef algorithm
%
% Example:
% load datareal
% s = Bex_lsef(f,d,6)
% fn(s,:)
%
% See also Bfs_lsef
%
% (c) GRIMA-DCCUC, 2011
% http://grima.ing.puc.cl
function s = Bex_lsef(f,d,m)
op.m = 2*m; % 2*m features will be selected using SFS
op.s = 0.75; % only 75% of sample will be used
op.show = 1; % display results
op.b.name = 'fisher'; % SFS with Fisher
s1 = Bfs_balu(f,d,op); % index of selected features
X1 = f(:,s1); % preselected features
op.m = m; % m/2 features will be selected
op.show = 1; % display results
[s2,Y,th] = Bfs_lsef(X1,op); % Y is computed as A*th, where
% A = [X1(:,s2) ones(size(X1,1),1)]
Ypca = Bft_pca(X1,0.95); % PCA analysis
mse(Y-Ypca) % Y and Ypca are very similar.
T1 = X1(:,s2); % selected features (transformation
op.p = [];
ds1 = Bcl_lda(T1,d,T1,op);
p1 = Bev_performance(d,ds1); % performance with lsef
T2 = Bft_pca(X1,op.m); % transformed features using PCA
op.p = [];
ds2 = Bcl_lda(T2,d,T2,op);
p2 = Bev_performance(d,ds2); % performance with PCA
fprintf('Performance with lsef = %5.4f, with PCA = %5.4f\n',p1,p2)
s = s1(s2);
|
github
|
domingomery/Balu-master
|
Bfs_clean.m
|
.m
|
Balu-master/FeatureSelection/Bfs_clean.m
| 1,773 |
utf_8
|
13cedc8d94836230ab53973127f23964
|
% selec = Bfs_clean(X,show)
%
% Toolbox: Balu
% Feature selection cleaning.
%
% It eliminates constant features and correlated features.
%
% Input: X is the feature matrix.
% show = 1 displays results (default show=0)
% Output: selec is the indices of the selected features
%
% Example:
% load datareal
% s = Bfs_clean(f,1); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:); % list of feature names
% disp('Original:'); howis(f) % original set of features
% disp('Selected:'); howis(X) % set of selecte features
%
% D.Mery, PUC-DCC, Jul. 2009
% http://dmery.ing.puc.cl
function selec = Bfs_clean(X,show)
f = X;
if not(exist('show','var'))
show = 0;
end
nf = size(f,2);
p = 1:nf;
ip = zeros(nf,1);
% eliminating correlated features
warning off
C = abs(corrcoef(f));
warning on
[ii,jj] = find(C>0.99);
if (not(isempty(ii)))
for i=1:length(ii)
if (abs(ii(i)-jj(i))>0)
k = max([ii(i) jj(i)]);
t = find(p==k);
n = length(p);
if (not(isempty(t)))
if (t==1)
p = p(2:n);
else
if (t==n)
p = p(1:n-1);
else
p = p([1:t-1 t+1:n]);
end
end
end
end
end
end
ip(p) = 1;
% eliminating constant features
s = std(f);
ii = find(s<1e-8);
if not(isempty(ii))
ip(ii) = 0;
end
p = find(ip);
fc = f(:,p);
nc = size(fc,2);
if show
fprintf('Bfs_clean: number of features reduced from %d to %d.\n',nf,nc)
end
selec=p;
|
github
|
domingomery/Balu-master
|
Bfs_sfscorr.m
|
.m
|
Balu-master/FeatureSelection/Bfs_sfscorr.m
| 1,392 |
utf_8
|
42d6288b4c73a7cbcc34d92b3fa48c76
|
% [R,selec] = Bfa_sfscorr(f,d,m)
%
% Toolbox: Balu
% Sequential Forward Selection for features f according to measurment d.
% The algorithm searchs the linear combination of the features that
% best correlates with d.
% m features will be selected.
% R is the obtained correlation coefficient and selec are the number of the
% selected features.
%
% D.Mery, PUC-DCC, Aug. 2008
% http://dmery.ing.puc.cl
%
function [R,selec] = Bfa_sfscorr(f,d,m)
selec = []; %selected features
R = [];
Rmax = 0;
k=1;
while (k<=m)
nuevo = 0;
for i=1:size(f,2)
if (k==1) || (sum(selec==i)==0)
if std(f(:,i))>0
s = [selec i];
fs = f(:,s);
[per,Rs] = Bfa_corrsearch(fs,d,1,2,0);
if (Rs>Rmax)
ks = i;
Rmax = Rs;
nuevo = 1;
end
end
end
end
if (nuevo)
selec = [selec ks];
R = [R Rmax];
clf
bar(R)
hold on
for i=1:length(selec)
text(i-0.4,R(i)*1.05,sprintf('%d',selec(i)));
end
pause(0)
k = k + 1;
else
disp('no more improvement, the sequantial search is interrupted.');
k = 1e10;
end
end
figure
Bfa_corrsearch(f(:,selec),d,1,2,1);
|
github
|
domingomery/Balu-master
|
Bfs_noposition.m
|
.m
|
Balu-master/FeatureSelection/Bfs_noposition.m
| 1,380 |
utf_8
|
6d584f9b6a2290b942e2363403888915
|
% [f_new,fn_new] = Bfs_noposition(f,fn)
%
% Toolbox: Balu
% This procedure deletes the features related to the position.
% It deletes the following features:
% - center of grav i
% - center of grav j
% - Ellipse-centre i
% - Ellipse-centre j
%
% Example:
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X1,Xn1] = Bfx_basicgeo(R); % basic geometric features
% [X2,Xn2] = Bfx_fitellipse(R); % Ellipse features
% X3 = [X1 X2]; Xn3 =[Xn1;Xn2];
% fprintf('\nOriginal features\n');
% Bio_printfeatures(X3,Xn3)
% [X4,Xn4] = Bfs_noposition(X3,Xn3); % delete position features
% fprintf('\nSelected features\n');
% Bio_printfeatures(X4,Xn4)
%
% See also Bfs_norotation, Bfs_nobackground.
%
% D.Mery, PUC-DCC, Nov. 2009
% http://dmery.ing.puc.cl
function [f_new,fn_new] = Bfs_noposition(f,fn)
s = ['center of grav i [px] '
'center of grav j [px] '
'Ellipse-centre i [px] '
'Ellipse-centre j [px] '];
[n,m] = size(s);
M = size(fn,1);
f_new = f;
fn_new = fn;
ii = [];
for i=1:n;
D = sum(abs(fn(:,1:m)-ones(M,1)*s(i,:)),2);
ii = [ii; find(D==0)];
end
if not(isempty(ii))
f_new(:,ii) = [];
fn_new(ii,:) = [];
end
|
github
|
domingomery/Balu-master
|
Bfs_fosmod.m
|
.m
|
Balu-master/FeatureSelection/Bfs_fosmod.m
| 3,173 |
utf_8
|
459b2c4292d50ca2234e5e3f78eda0fc
|
% selec = Bfs_fosmod(X,options)
%
% Toolbox: Balu
% Feature Selection using FOS-MOD algorithm
%
% input: X feature matrix
% options.m number of features to be selected
% options.show = 1 displays results
%
% output: selec selected features
%
% FOS_MOD is a forward orthogonal search (FOS) algorithm by maximizing
% the overall dependency (MOD). The algorithm assumes that a linear
% relationship exists between sample features:
%
% Reference:
% Wei, H.-L. & Billings, S. Feature Subset Selection and Ranking for
% Data Dimensionality Reduction Pattern Analysis and Machine
% Intelligence, IEEE Transactions on, 2007, 29, 162-166
%
% Example:
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 5; % 5 features will be selected
% op.show = 1; % display results
% s = Bfs_fosmod(X1,op); % index of selected features
% T1 = X1(:,s); % selected features (transformation
% % is avoided)
% op.p = [];
% ds1 = Bcl_lda(T1,d,T1,op);
% p1 = Bev_performance(d,ds1) % performance with fosmod
% T2 = Bft_pca(X1,op.m); % transformed features using PCA
% op.p = [];
% ds2 = Bcl_lda(T2,d,T2,op);
% p2 = Bev_performance(d,ds2) % performance with PCA
% fprintf('Performance with fosmod = %5.4f, with PCA = %5.4f\n',p1,p2)
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function selec = Bfs_fosmod(X,d,options)
% d is not used by this algorithm.
% the sintaxis "Bfs_fosmod(X,d,options)" is allowed in order to be
% similar to other Bfs_ functions.
if nargin==2;
options = d;
end
m = options.m;
show = options.show;
[N,n] = size(X); % number of instances (samples), number of features
l = zeros(m,1);
p = ones(n,1);
q = zeros(N,n);
for mm=1:m
if mm==1
Q = X;
else
Q = zeros(N,n);
for j=1:n
if p(j)
alj = X(:,j);
qm = alj;
for k=1:mm-1
qm = qm - ((alj'*q(:,k))/(q(:,k)'*q(:,k)))*q(:,k);
end
Q(:,j) = qm;
end
end
end
C = zeros(n,n);
for j=1:n
if p(j)
for i=1:n
C(i,j) = Bfa_sqcorrcoef(X(:,i),Q(:,j));
end
end
end
Cm = mean(C);
[i,l(mm)] = max(Cm);
p(l(mm)) = 0;
q(:,mm) = Q(:,l(mm));
err = zeros(n,mm);
for k=1:mm
for j=1:n
err(j,k) = Bfa_sqcorrcoef(X(:,j),q(:,k))*100;
end
end
if mm>1
serr = sum(err,2);
else
serr = err;
end
mserr = mean(serr);
if show
fprintf('%2d) selected feature=%4d error=%7.2f%%\n',mm,l(mm),100-mserr)
end
end
selec = l;
|
github
|
domingomery/Balu-master
|
Bfs_random.m
|
.m
|
Balu-master/FeatureSelection/Bfs_random.m
| 2,187 |
utf_8
|
99545454631b8da08b171978d75886f7
|
% selec = Bfs_random(X,d,options)
%
% Toolbox: Balu
% Select best features from random subsets of X according to ideal
% classification d.
% options.m is the number features will be selected.
% options.M is the number of random subsets to be tested.
% options.b.method = 'fisher' uses Fisher objetctive function.
% options.b.method = 'sp100' uses as criteria Sp @Sn=100%.
% options.b can be any Balu classifier structure (see example).
% options.show = 1 display results.
% selec is the indices of the selected features.
%
% Example 1: Feature selection using Fisher dicriminant
% load datareal
% op.m = 10; % 10 features will be selected
% op.M = 1000; % number of random sets to be tested
% op.show = 1; % display results
% op.b.name = 'fisher'; % Feature score
% s = Bfs_random(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% Example 2: Feature selection using a KNN classifier
% load datareal
% op.m = 10; % 10 features will be selected
% op.M = 1000; % number of random sets to be tested
% op.show = 1; % display results
% op.b.name = 'knn'; % Feature score is the performance
% op.b.options.k = 5; % 5 neighbors
% s = Bfs_random(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% (c) D.Mery, PUC-DCC, Jul. 2011
% http://dmery.ing.puc.cl
function selec = Bfs_random(X,d,options)
m = options.m;
M = options.M;
show = options.show;
s0 = Bfs_clean(X);
f = X(:,s0);
nf = size(f,2);
Jmax = 0;
k = 0;
while (k<=M)
N = rand(nf,1);
[i,j] = sort(N);
s = j(1:m);
fs = f(:,s);
Js = Bfa_score(fs,d,options);
if (Js>Jmax)
selec = s;
Jmax = Js;
if show
fprintf('Jmax = %8.4f\n',Jmax);
end
end
k = k+1;
end
|
github
|
domingomery/Balu-master
|
Bfs_norotation.m
|
.m
|
Balu-master/FeatureSelection/Bfs_norotation.m
| 1,537 |
utf_8
|
b1c6fa0ea43372e402874e2f3e6db857
|
% [f_new,fn_new] = Bfs_norotation(f,fn)
%
% Toolbox: Balu
% This procedure deletes all no rotation invariant features.
% It deletes the features that have in their name the strings:
% - orient
% - Gabor(
% - [8,u2] for LBP
% - sLBP
%
% Example:
% options.b = Bfx_build({'haralick','lbp'});
% options.colstr = 'rgb'; % R image
% I = imread('testimg1.jpg'); % input image
% R = Bim_segbalu(I); % segmentation
% [X,Xn] = Bfx_int(I,R,options); % intensity features
% [X1,Xn1] = Bfs_norotation(X,Xn);
% Bio_printfeatures(X1,Xn1)
%
%
% See also Bfs_noposition, Bfs_nobackground.
%
% D.Mery, PUC-DCC, May 2012
% http://dmery.ing.puc.cl
function [f_new,fn_new] = Bfs_norotation(f,fn)
f_new = f; fn_new = fn;
[ix,fn_new] = Bio_findex(fn_new,'Orientation',0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'Ellipse-orient',0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'Gabor(',0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'[8,u2]',0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'sLBP' ,0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'Fourier Abs (' ,0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'Fourier Ang (' ,0); f_new = f_new(:,ix);
[ix,fn_new] = Bio_findex(fn_new,'DCT(' ,0); f_new = f_new(:,ix);
|
github
|
domingomery/Balu-master
|
Bfs_all.m
|
.m
|
Balu-master/FeatureSelection/Bfs_all.m
| 131 |
utf_8
|
844e98d7bb621a3acffe6f60df432c78
|
% Dummy file called by Bfx_gui
% It selects all features of X.
function selec = Bfs_all(X,d,options)
m = size(X,2);
selec = (1:m)';
|
github
|
domingomery/Balu-master
|
Bfs_balu.m
|
.m
|
Balu-master/FeatureSelection/Bfs_balu.m
| 2,884 |
utf_8
|
553453caae55239bed210fd8551d179a
|
% selec = Bfs_balu(X,d,options)
%
% Toolbox: Balu
% Feature selection of "best" options.m features of X to ideal
% classification d. This function uses only a portion of options.s
% samples to select the features.
% Bfs_balu (1) normalizes (using Bft_norm), (2) cleans (using Bfs_clean)
% and (3) selects features (using Bfs_sfs).
% options.b.method = 'fisher' uses Fisher objetctive function.
% options.b.method = 'sp100' uses as criteria Sp @Sn=100%.
% options.b can be any Balu classifier structure (see example).
% options.show = 1 display results.
% selec is the indices of the selected features.
%
% Example 1: Balu feature selection using Fisher dicriminant
% load datareal
% op.m = 10; % 10 features will be selected
% op.s = 0.75; % only 75% of sample will be used
% op.show = 1; % display results
% op.b.name = 'fisher'; % SFS with Fisher
% s = Bfs_balu(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% Example 2: Balu feature selection using a KNN classifier
% load datareal
% op.m = 10; % 10 features will be selected
% op.s = 0.75; % only 75% of sample will be used
% op.show = 1; % display results
% op.b.name = 'knn'; % SFS with KNN
% op.b.options.k = 5; % 5 neighbors
% s = Bfs_balu(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function selec = Bfs_balu(X,d,options)
s = options.s;
show = options.show;
m = options.m;
if (s<0)||(s>1)
error('Bfs_balu: options.s must be between 0 and 1...')
end
fT = X;
dT = d;
if show
close all
end
if s<1
[f,d] = Bds_stratify(fT,dT,s);
else
f = X;
end
% cleaning
if show
disp('eliminating no relevant and correlated features...');
end
selec0 = Bfs_clean(f);
f = f(:,selec0);
% normalize
if show
disp('normalizing features...');
end
ff = Bft_norm(f,1);
% selection
if show
disp('selecting features...');
end
M = size(ff,2);
if M<m
error('Bfs_balu: number of features to be selected (%d) is larger than number of existing features (%d)',M,m);
end
warning off %#ok<WNOFF>
N = nchoosek(M,m);
warning on %#ok<WNON>
if N>10000
if show
disp('selecting features using SFS...');
end
selec = Bfs_sfs(ff,d,options);
else
if show
disp('selecting features using exhaustive search...');
end
selec = Bfs_exsearch(ff,d,options);
end
selec = selec0(selec);
|
github
|
domingomery/Balu-master
|
Bfs_mRMR.m
|
.m
|
Balu-master/FeatureSelection/Bfs_mRMR.m
| 4,499 |
utf_8
|
e8736538d2bdd540fea3d416d8e0dee6
|
% selec = Bfs_mRMR(X,d,options)
%
% Toolbox: Balu
% Feature selection using Criteria of Max-Dependency, Max-Relevance, and
% Min-Redundancy after Peng et al. (2005)
%
% X extracted features (NxM): N samples, M features
% d ideal classification (Nx!) for the N samples
% options.m number of selected features
% options.p a priori probability of each class (if not given, is
% estimated using the ratio samples per class to N.
%
% References:
% Peng, H.; Long, F.; Ding, C. (2005): Feature Selection Based on Mutual
% Information: Criteria of Max-Dependency, Max-Relevance, and Min-
% Redundancy. IEEE Trans. on Pattern Analysis and Machine Intelligence,
% 27(8):1226-1238.
%
% Z. I. Botev, J. F. Grotowski, and D. P. Kroese (2010): Kernel density
% estimation via diffusion Annals of Statistics, 38(5):2916-2957.
%
% NOTE:
% The pdf's are estimated using Kernel Density Estimations programs
% kde.m and kde2d.m after Botev et al. (2010) implemented by Botev.
% These files are in Balu directory 'Feature Analysis' as Bfa_kde and
% Bfs_kde2d. They can also be downloaded from www.mathwork.com
% (c) Zdravko Botev. All rights reserved.
%
% Example (comparison between SFS-Fisher and mRMR):
%
% load datareal
% s = Bfs_clean(f,1);
% X = f(:,s);
% k = 0;
% k=k+1;b(k).name = 'knn'; b(k).options.k = 9; % KNN with 9 neighbors
% k=k+1;b(k).name = 'lda'; b(k).options.p = []; % LDA
% k=k+1;b(k).name = 'qda'; b(k).options.p = []; % QDA
% k=k+1;b(k).name = 'dmin'; b(k).options = []; % Euclidean distance
% k=k+1;b(k).name = 'maha'; b(k).options = []; % Mahalanobis distance
% k=k+1;b(k).name = 'libsvm'; b(k).options.kernel = '-t 2'; % rbf-SVM
% k=k+1;b(k).name = 'nnglm'; b(k).options.method = 3; b(k).options.iter = 10; % Nueral network
%
% op.strat=1; op.b = b; op.v = 10; op.show = 1; op.c = 0.95; % 10 groups cross-validation
%
% msel1 = 12;
% msel2 = 4;
%
% op1.m = msel1;op1.show=1;op1.b.name='fisher';
% s1 = Bfs_sfs(X,d,op1);
% X1 = X(:,s1);
% disp('Performances using SFS-Fisher:')
% p1 = Bev_crossval(X1(:,1:msel2),d,op);
%
% op2.m = msel2;op2.show=1;
% s2 = Bfs_mRMR(X1,d,op2);
% disp(' ')
% disp('Performances using SFS-mRMR:')
% p2 = Bev_crossval(X1(:,s2),d,op);
% disp(' ')
% disp('Comparison of performances and mean of performances:')
% [p1 p2], mean([p1 p2])
%
% D.Mery, PUC-DCC, Jul. 2011
% http://dmery.ing.puc.cl
function s = Bfs_mRMR(X,d,options)
if ~isfield(options,'p')
dn = max(d)-min(d)+1; % number of classes
p = ones(dn,1)/dn;
else
p = options.p;
end
if ~isfield(options,'show')
show = 0;
else
show = options.show;
end
% T = 100;
M = size(X,2);
n = options.p;
s = zeros(n,1);
t = ones(M,1);
h = zeros(n,1);
ff = Bio_statusbar('Bfs_mRMR');
Ic = zeros(M,1);
m = 1;
for j=1:M
% Ic(j) = Bfa_mutualinfo(X(:,j),d,p); % see in paper I(xj;c)
Ic(j) = Bfa_miparzen2([X(:,j) d],1,p); % see in paper I(xj;c)
end
ff = Bio_statusbar(1/n,ff);
[Icmax,jsel] = max(Ic);
if show
clf
h(m) = Icmax;
bar(h);
end
s(m) = jsel;
t(jsel) = 0;
MI = zeros(M,M);
if n>1
for m=2:n
Jmax = -Inf;
for j=1:M
if t(j)==1 % if yes, feature j is not selected
xj = X(:,j);
sumI = 0;
for i=1:m-1
xi = X(:,s(i));
% sumI = sumI + Bfa_mutualinfo2([xj xi]); % see in paper I(xj;xi)
if MI(s(i),j)==0;
mij = Bfa_miparzen2([xj xi]);
MI(s(i),j) = mij;
MI(j,s(i)) = mij;
else
mij = MI(s(i),j);
end
sumI = sumI + mij; % see in paper I(xj;xi)
% sumI = sumI + Bfa_miparzen2([xj xi]); % see in paper I(xj;xi)
end
% Jj = Ic(j)-sumI/(m-1);
Jj = Ic(j)/(sumI/(m-1));
% Jj = Ic(j)-sumI;
if Jj>Jmax
Jmax = Jj;
jsel = j;
end
end
end
s(m) = jsel;
t(jsel) = 0;
if show
h(m) = Jmax;
bar(h)
end
ff = Bio_statusbar(m/n,ff);
end
end
delete(ff)
|
github
|
domingomery/Balu-master
|
Bfs_bb.m
|
.m
|
Balu-master/FeatureSelection/Bfs_bb.m
| 4,988 |
utf_8
|
149d76a182777b35b877fa7080dc39a7
|
% selec = Bfs_bb(X,d,options)
%
% Toolbox: Balu
% Feature selection using Branch & Bound for fatures X according to
% ideal classification d. optins.m features will be selected.
% options.b.method = 'fisher' uses Fisher objetctive function.
% options.b.method = 'sp100' uses as criteria Sp @Sn=100%.
% options.b can be any Balu classifier structure (see example).
% options.show = 1 display results.
% selec is the indices of the selected features.
%
% Example 1: Branch and Bound search from using Fisher dicriminant
% (compare this example with exhaustive serach)
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 3; % 3 features will be selected
% op.show = 1; % display results
% op.b.name = 'fisher'; % SFS with Fisher
% s = Bfs_bb(X1,d,op); % index of selected features
% X2 = X1(:,s); % selected features
% Xn2 = Xn1(s,:) % list of feature names
%
% Example 2: Branch and Bound using a KNN classifier
% (compare this example with exhaustive serach)
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 4; % 3 features will be selected
% op.show = 1; % display results
% op.b.name = 'knn'; % SFS with KNN
% op.b.options.k = 5; % 5 neighbors
% s = Bfs_bb(X1,d,op); % index of selected features
% X2 = X1(:,s); % selected features
% Xn2 = Xn1(s,:) % list of feature names
%
% (c) Irene Zuccar, PUC-DCC, Jul. 2011
% http://grima.ing.puc.cl
function selec = Bfs_bb(X,d,options)
f = X;
p = options.m;
NumFeat = size(f,2);
Bend = false;
% Initial node in Bnode
Bnode = nchoosek(1:NumFeat,NumFeat);
% Score for initial node
JBnode = Bfa_score(f,d,options);
if isnan(JBnode)
Bnode = [1,Bnode];
else
Bnode = [JBnode,Bnode];
end
% Add initial node to stack Bopened.
Bopened = [];
Bclosed = [];
Bopened = [Bnode;Bopened];
bestJ = 0;
selec = [];
r = 0;
while (~Bend)
% Take Bnode from stack and store in Bclosed
[Bopened,Bclosed,Bnode] = Bfs_bbpop(Bopened,Bclosed);
JBnode=Bnode(1,1);
% Prune criteria
if JBnode > bestJ
b = size(Bnode,2)-1;
% If it has not reached the leaves then create the children of Bnode
if b>p
% It generates the children that have not already generated,
% it computes the score and it stores and sorts in Bopened,
% it takes the best one.
h = Bfs_bbchild(Bnode,f,d,Bopened,Bclosed,options);
Bopened = Bfs_bbpush(Bopened,h);
end
% If Bnode is a leave > update prune criteria, and store the
% selected features
if b == p
bestJ = JBnode;
selec = Bnode;
end
end
% If empty stack then end search in tree
a = size(Bopened,1);
if a == 0
Bend = true;
end
% The best result in 10000 iterations
r=r+1;
if r==10000
Bend = true;
error('Branch and bound for feature selection interrupted: too many iterations.')
end
end
selec = selec(2:end)';
end
function resp = Bfs_bbis(Bnode,Bopened,Bclosed)
resp = false;
m = size(Bopened,1);
for i = 1:m
if Bopened(i,2:end) == Bnode(2:end)
resp = true;
return;
end
end
m = size(Bclosed,1);
for i = 1:m
if Bclosed(i,2:end) == Bnode(2:end)
resp=true;
return;
end
end
end
function resp = Bfs_bbchild(padre,f,d,Bopened,Bclosed,options)
NumFeat = size(Bclosed,2)-1;
nPadre = size(padre,2)-1;
Bchild = nchoosek(padre(1,2:nPadre+1),nPadre-1);
Bchild = [zeros(nPadre,1),Bchild];
Bchild = padarray(Bchild,[0,NumFeat-nPadre+1],0,'post');
resp=[];
for i=1:nPadre
if ~Bfs_bbis(Bchild(i,:),Bopened,Bclosed)
fchild = f(:,Bchild(i,2:nPadre));
% JBchild = Bfa_jfisher(fchild,d);
JBchild = Bfa_score(fchild,d,options);
Bchild(i,1) = JBchild;
resp = [Bchild(i,:);resp];
end
end
resp = -sortrows(-resp);
end
function [Bopened,Bclosed,Bnode] = Bfs_bbpop(Bopened,Bclosed)
Bnode = Bopened(1,:);
Bclosed = [Bnode;Bclosed];
aux=Bnode(2:end);
pos = find(aux==0);
q = size(pos,2);
if q~=0
Bnode(pos+1:end) = [];
end
Bopened(1,:) = [];
end
function Bopened = Bfs_bbpush(Bopened,h)
Bopened = [h;Bopened];
end
|
github
|
domingomery/Balu-master
|
Bfs_nobackground.m
|
.m
|
Balu-master/FeatureSelection/Bfs_nobackground.m
| 1,499 |
utf_8
|
98cdc4bfb82c9a67adee6e5799d28faa
|
% [f_new,fn_new] = Bfs_nobackground(f,fn)
%
% Toolbox: Balu
% This procedure deletes the features related to the position.
% It deletes the following features related to the contrast:
% - contrast-K1
% - contrast-K2
% - contrast-K3
% - contrast-Ks
% - contrast-K
%
% Example:
% options.show = 1; % display results
% options.neighbor = 2; % neigborhood is imdilate
% options.param = 5; % with 5x5 mask
% I = imread('testimg4.jpg'); % input image
% J = I(395:425,415:442,1); % region of interest (red)
% R = J<=130; % segmentation
% figure;imshow(J,[])
% figure;imshow(R)
% [X1,Xn1] = Bfx_contrast(J,R,options); % contrast features
% options.mask = 5; % display results
% [X2,Xn2] = Bfx_basicint(J,R,options);
% X3 = [X1 X2]; Xn3 =[Xn1;Xn2];
% fprintf('\nOriginal features\n');
% Bio_printfeatures(X3,Xn3)
% [X4,Xn4] = Bfs_nobackground(X3,Xn3); % delete contrast features
% fprintf('\nSelected features\n');
% Bio_printfeatures(X4,Xn4)
%
% D.Mery, PUC-DCC, May 2012
% http://dmery.ing.puc.cl
function [f_new,fn_new] = Bfs_nobackground(f,fn)
[ix,fn_new] = Bio_findex(fn,'contrast',0);
f_new = f(:,ix);
|
github
|
domingomery/Balu-master
|
Bfs_exsearch.m
|
.m
|
Balu-master/FeatureSelection/Bfs_exsearch.m
| 2,602 |
utf_8
|
74412cf1479a6c06be68936e449dcd5a
|
% selec = Bfs_exsearch(X,d,options)
%
% Toolbox: Balu
% Feature selection using exhaustive search for fatures X according to
% ideal classification d. optins.m features will be selected.
% options.b.method = 'fisher' uses Fisher objetctive function.
% options.b.method = 'sp100' uses as criteria Sp @Sn=100%.
% options.b can be any Balu classifier structure (see example).
% options.show = 1 display results.
% selec is the indices of the selected features.
%
% Example 1: Exhaustive search from using Fisher dicriminant
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 3; % 3 features will be selected
% op.show = 1; % display results
% op.b.name = 'fisher'; % SFS with Fisher
% s = Bfs_exsearch(X1,d,op); % index of selected features
% X2 = X1(:,s); % selected features
% Xn2 = Xn1(s,:) % list of feature names
%
% Example 2: Exhaustive serach using a KNN classifier
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 4; % 4 features will be selected
% op.show = 1; % display results
% op.b.name = 'knn'; % Feature selection using KNN
% op.b.options.k = 5; % 5 neighbors
% s = Bfs_exsearch(X1,d,op); % index of selected features
% X2 = X1(:,s); % selected features
% Xn2 = Xn1(s,:) % list of feature names
%
% D.Mery, PUC-DCC, Jul. 2011
% http://dmery.ing.puc.cl
function selec = Bfs_exsearch(X,d,options)
m = options.m;
show = options.show;
f = X;
M = size(f,2);
N = nchoosek(M,m);
if (N>10000)
ok = input(sprintf('Exhaustive Search needs %d evaluations... continue [yes=1/no=0]?',N));
if not(ok)
error('Exhaustive search for feature selection interrupted.')
end
end
T = nchoosek(1:M,m);
Jmax = 0;
for i=1:N
fs = f(:,T(i,:));
Js = Bfa_score(fs,d,options);
if (Js>Jmax)
Jmax = Js;
ks = i;
if show
fprintf('step=%2d/%d J=%f\n',i,N,Jmax)
end
end
end
selec = T(ks,:)';
|
github
|
domingomery/Balu-master
|
Bfs_ransac.m
|
.m
|
Balu-master/FeatureSelection/Bfs_ransac.m
| 1,759 |
utf_8
|
f416554f1291419b4b55cfb98cc3c01b
|
% selec = Bfsransac(X,d,m,show,method,param,param2,param3)
%
% Toolbox: Balu
% Sequential Forward Selection for fatures X according to ideal
% classification d. m features will be selected.
% method = 'fisher' uses Fisher objetctive function (in this case param
% is the a priori probability of each class, if not given it will be
% assumed constant).
% method = 'sp100' uses as criteria Sp @Sn=100%.
% method could be any classifier implemented in Balu with the
% corresponding parameters, e.g., method = 'knn' and param = 10, it will
% select the best m features for 10 nearest neighbours.
% show = 1 display results.
% selec is the indices of the selected features.
%
% D.Mery, PUC-DCC, Ago. 2010
% http://dmery.ing.puc.cl
%
function selec = Bfsransac(X,d,m,show,method,varargin)
s0 = Bfsclean(X);
f = X(:,s0);
nf = size(f,2);
selec = []; %selected features
Jmax = 0;
dn = max(d)-min(d)+1; % number of classes
if not(exist('show','var'))
show = 0;
end
if not(exist('method','var'))
method = 'fisher';
end
k = 0;
while (k<=300)
[f1,d1] = Bstratify(f,d,0.67);
s = Bfsfs(f1,d1,m);
fs = f(:,s);
switch lower(method)
case 'fisher'
if (not(exist('param','var')))
p = ones(dn,1)/dn;
end
Js = jfisher(fs,d,p);
case 'sp100'
Js = sp100(fs,d);
otherwise
e = ['ds = ' method '(fs,d,fs,varargin{:});'];
eval(e);
Js = Bperformance(d,ds);
end
if (Js>Jmax)
selec = s;
Jmax = Js;
if show
fprintf('Jmax = %8.4f\n',Jmax);
end
end
k = k+1;
end
|
github
|
domingomery/Balu-master
|
Bfs_sfs.m
|
.m
|
Balu-master/FeatureSelection/Bfs_sfs.m
| 3,719 |
utf_8
|
f9620a7255abdf0e8d1fe5a3fe189ce1
|
% selec = Bfs_sfs(X,d,options)
%
% Toolbox: Balu
% Sequential Forward Selection for fatures X according to ideal
% classification d. optins.m features will be selected.
% options.b.method = 'fisher' uses Fisher objetctive function.
% options.b.method = 'sp100' uses as criteria Sp @Sn=100%.
% options.b can be any Balu classifier structure (see example).
% options.show = 1 display results.
% selec is the indices of the selected features.
%
% Example 1: SFS using Fisher dicriminant
% load datareal
% op.m = 10; % 10 features will be selected
% op.show = 1; % display results
% op.b.name = 'fisher'; % SFS with Fisher
% s = Bfs_sfs(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
% op_lda.p = [];
% ds = Bcl_lda(X,d,X,op_lda); % LDA classifier
% p = Bev_performance(d,ds) % performance with sfs
%
% Example 2: SFS using a KNN classifier
% load datareal
% op.m = 10; % 10 features will be selected
% op.show = 1; % display results
% op.b.name = 'knn'; % SFS with KNN
% op.b.options.k = 5; % 5 neighbors
% s = Bfs_sfs(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% Example 3: SFS using sp100 criterion
% load datareal
% op.m = 10; % 10 features will be selected
% op.show = 1; % display results
% op.b.name = 'sp100'; % SFS with sp100 criterion
% s = Bfs_sfs(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% (c) D.Mery, PUC-DCC, Jul. 2011
% http://dmery.ing.puc.cl
function selec = Bfs_sfs(X,d,options)
m = options.m;
show = options.show;
if ~isfield(options,'force')
options.force = 0;
end
force = options.force;
f = X;
N = size(f,2);
selec = []; %selected features
J = zeros(m,1);
% Jmax = 0;
k = 0;
if show
ff = Bio_statusbar('SFS progress');
end
while (k<m)
if show
ff = Bio_statusbar(k/m,ff);
end
fnew = 0;
Jmax = -Inf;
for i=1:N
if (k==0) || (sum(selec==i)==0)
s = [selec; i];
fs = f(:,s);
Js = Bfa_score(fs,d,options);
if (Js>=Jmax)
ks = i;
Jmax = Js;
fnew = 1;
end
end
end
if (fnew)
selec = [selec; ks];
k = k + 1;
J(k) = Jmax;
if show
clf
bar(J)
fprintf('Jmax = %8.4f\n',Jmax);
hold on
for i=1:length(selec)
text(i-0.4,J(i)*1.05,sprintf('%d',selec(i)));
end
pause(0)
end
else
disp('Bfs_sfs: no more improvement. Sequential search for feature selection is interrupted.');
if and(force,(k<m))
fprintf('Bfs_sfs: Warning! %d random features were selected in order to have\n',m-k);
fprintf(' %d selected features (options.force is true).\n',m);
t = 1:N;
t(selec) = [];
n = length(t);
x = rand(n,1);
[i,j] = sort(x);
selec = [selec; t(j(1:m-k))' ];
end
k = 1e10;
end
end
if show
delete(ff);
end
|
github
|
domingomery/Balu-master
|
Bfs_rank.m
|
.m
|
Balu-master/FeatureSelection/Bfs_rank.m
| 2,709 |
utf_8
|
04c850f50a535397c9a1dc6eaaf5ce03
|
% selec = Bfs_rank(X,d,options)%
%
% Toolbox: Balu
% Feature selection based on command rankfeatures (from MATLAB
% Bioinformatics Toolbox) that ranks ranks key features by class
% separability criteria.
%
% input: X feature matrix
% options.m number of features to be selected
% options.criterion can be:
% 'ttest' (default) Absolute value two-sample T-test with pooled
% variance estimate
% 'entropy' Relative entropy, also known as Kullback-Lieber
% distance or divergence
% 'brattacharyya' Minimum attainable classification error or
% Chernoff bound
% 'roc' Area between the empirical receiver operating
% characteristic (ROC) curve and the random classifier
% slope
% 'wilcoxon' Absolute value of the u-statistic of a two-sample
% unpaired Wilcoxon test, also known as Mann-Whitney
%
% Notes: 1) 'ttest', 'entropy', and 'brattacharyya' assume normal
% distributed classes while 'roc' and 'wilcoxon' are nonparametric tests,
% 2) all tests are feature independent.
%
% output: selec selected features
%
% Example:
% load datareal
% op.m = 10; % 10 features will be selected
% op.criterion = 'roc'; % ROC criterion will be used
% op.show = 1; % display results
% s = Bfs_rank(f,d,op); % index of selected features
% X = f(:,s); % selected features
% Xn = fn(s,:) % list of feature names
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
%
function selec = Bfs_rank(X,d,options)
m = options.m;
criterion = options.criterion;
dmin = min(d);
dmax = max(d);
k = dmax-dmin+1;
if not(exist('criterion','var'))
criterion = 'ttest';
end
if k<2
error('Bfs_rank: Number of classes of d must be greater than 1.');
end
if ~exist('rankfeatures','file')
error('Bfs_rank: This function requires Bioinformatics Toolbox.');
end
if k==2;
idx = rankfeatures(X',d,'criterion',criterion);
selec = idx(1:m);
else
[N,M] = size(X);
S = zeros(M,k);
for j=1:k
dj = (d==j)+1;
S(:,j) = rankfeatures(X',dj,'criterion',criterion);
end
T = S';
idx = T(:);
i = 1;
selec = zeros(m,1);
selec(1) = idx(1);
j=2;
while i<=m
if not(ismember(idx(j),selec))
selec(i) = idx(j);
i = i+1;
end
j = j + 1;
end
end
|
github
|
domingomery/Balu-master
|
Bfs_lsef.m
|
.m
|
Balu-master/FeatureSelection/Bfs_lsef.m
| 4,224 |
utf_8
|
43b4356a3d4d8fcb5317ebef270f1153
|
% [selec,Y,th] = Bfs_lsef(X,options)
%
% Toolbox: Balu
% Feature Selection using LSE-forward algorithm
%
% input: X feature matrix
% options.m number of features to be selected
% optoins.show = 1 displays results
%
% output: selec selected features
% Y is equal to A*th, where A = [X(:,selec) ones(size(X,1),1)]
% Y is very similar to PCA.
%
% The main idea of the algorithms is to evaluate and select feature
% subsets based on their capacities to reproduce sample projections on
% principal axes:
%
% Reference:
% Mao, K. Identifying critical variables of principal components for
% unsupervised feature selection Systems, Man, and Cybernetics, Part B:
% Cybernetics, IEEE Transactions on, 2005, 35, 339-344
%
% Example 1: using PCA
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 6; % 5 features will be selected
% op.show = 1; % display results
% op.pca = 1;
% s = Bfs_lsef(X1,d,op); % index of selected features
% T1 = X1(:,s); % selected features (transformation
% % is avoided)
% op.p = [];
% ds1 = Bcl_lda(T1,d,T1,op);
% p1 = Bev_performance(d,ds1) % performance with lsef
% T2 = Bft_pca(X1,op.m); % transformed features using PCA
% op.p = [];
% ds2 = Bcl_lda(T2,d,T2,op);
% p2 = Bev_performance(d,ds2) % performance with PCA
% fprintf('Performance with lsef = %5.4f, with PCA = %5.4f\n',p1,p2)
%
% Example 2: Linear transformation of selected features is similar
% to PCA.
% load datareal
% s1 = [279 235 268 230 175 165 207 160 269 157]; %indices using Example
% %of Bfs_sfs.
% X1 = f(:,s1); % preselected features
% Xn1 = fn(s1,:);
% op.m = 6; % 6 features will be selected
% op.show = 1; % display results
% op.pca = 1;
% [s,Y,th] = Bfs_lsef(X1,d,op); % Y is computed as A*th, where
% % A = [X1(:,s) ones(size(X1,1),1)]
% Yt = Bft_pca(X1,op.m); % PCA analysis
% sqdif(Y,Yt) % Y and Yt are very similar.
%
% See also Bft_lseft, Bft_pca, Bft_plsr.
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function [selec,Y,th] = Bfs_lsef(X,d,options)
% d is not used by this algorithm.
% the sintaxis "Bfs_fosmod(X,d,options)" is allowed in order to be
% similar to other Bfs_ functions.
if nargin==2;
options = d;
end
m = options.m;
show = options.show;
if isfield(options,'pca')
pcat = options.pca;
else
pcat = 0;
end
[N,n] = size(X); % number of instances (samples), number of features
% 1) Initialize S to an empty set
% S = [];
% 2) Initialize R to the full feature set
% R = X;
% 3) Perform PCA on complete data:
% 4) Calaculate sample projections on the first principal axes with 0.95
% energy
if pcat
Yt = Bft_pca(X,m);
else
Yt = Bft_plsr(X,d,m);
end
Ny = length(Yt(:));
% 5) m iterations for m features
p = ones(n,1); % '1' means not selected feature
for l=1:m
em = Inf*ones(n,1);
S = X(:,not(p));
Yss = zeros(N,Ny/N,n);
ths = zeros(l+1,Ny/N,n);
for j=1:n
if p(j)
A = [S X(:,j) ones(N,1)];
th = (A'*A)\A'*Yt;
ths(:,:,j) = th;
Ys = A*th;
em(j) = norm(Yt-Ys);
Yss(:,:,j) = Ys;
end
end
[emin,js] = min(em); % js is the feature that minimizes Yt-Ys, i.e.,
% Ys is the best fit of Yt computed as a linear
% transformation of [S X(:,js) ones(N,1)]
Y = Yss(:,:,js);
th = ths(:,:,js);
if show
fprintf('%2d) selected feature=%4d error=%7.2f%%\n',l,js,emin/Ny*100)
end
p(js) = 0;
end
selec = find(p==0);
|
github
|
domingomery/Balu-master
|
Bmv_epidist.m
|
.m
|
Balu-master/MultiView/Bmv_epidist.m
| 1,671 |
utf_8
|
23fe88b6fd0a2d7bb2ba85a4e316b8ed
|
% d = Bmv_epidist(m1,m2,F,method)
%
% Toolbox: Balu
%
% Distance from m2 to epipolar line l2 = F*m1
%
% d = distance2(m1,m2,F,'method') returns the distance
% error. The posible corresponding points are m2 and m1.
% F is the fundamental matrix. The distance is calculated
% using the following methods:
%
% method = 'euclidean': uses Euclidean distance from m2 to l=F*m1.
%
% method = 'sampson': uses Sampson distance.
%
% If no method is given, 'euclidean' will be assumed as default.
%
% Both methods can be found in:
%
% R. Hartley and A. Zisserman. Multiple View Geometry in Computer
% Vision. Cambridge University Press, 2000.
%
% Example:
% A = rand(3,4); % Projection matrix for view 1
% B = rand(3,4); % Projection matrix for view 1
% M = [1 2 3 1]'; % 3D point (X=1,Y=2,Z=3)
% m1 = A*M;m1 = m1/m1(3); % projection point in view 1
% m2 = B*M;m2 = m2/m2(3); % projection point in view 2
% F = Bmv_fundamental(A,B,'tensor'); % Fundamental matrix using tensors
% d = Bmv_epidist(m1,m2,F,'sampson') % sampson distance to epipolar line
%
% See also Bmv_epiplot.
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function d = Bmv_epidist(m1,m2,F,method)
if ~exist('method','var')
method = 'euclidean';
end
d0 = abs(m2'*F*m1);
switch lower(method)
case 'sampson'
l1 = F*m1;
l2 = F'*m2;
d = d0/sqrt(l1(1)^2+l1(2)^2+l2(1)^2+l2(2)^2);
otherwise % euclidean
l = F*m1;
d = d0/sqrt(l(1)^2+l(2)^2);
end
|
github
|
domingomery/Balu-master
|
Bmv_reco3dna.m
|
.m
|
Balu-master/MultiView/Bmv_reco3dna.m
| 2,239 |
utf_8
|
29e34b59c3f3155e8f17b93b6b09344e
|
% [M,err,ms] = Bmv_reco3dna(m,P)
%
% Toolbox: Balu
%
% 3D affine reconstruction from n corresponding points
%
% It returns a 3D point M that fullfils
% the following projective equations:
%
% m1 = P1*M
% m2 = P2*M
% :
% where mk = m(:,k) are the 2D projection points of 3D point M
% in image k; Pk = P(k*3-2:k*3,:) the corresponding 3x4 projection. The
% last row of a affine projection matrix must be [0 0 0 1]. factors
% lambda k are 1 in affine projections.
%
% mk and M are given in homogeneous coordinates, i.e., mk
% are 3x1 vectors, and M is a 4x1 vector.
%
% ms is the reprojection of M in each view, ideally ms=m.
% err is the reprojection error calculated as norm of ms-m.
% The method used in this program is proposed in:
%
% R. Hartley. A linear method for reconstruction from lines and
% points. In 5th International Conference on Computer Vision
% (ICCV-95), pages 882-887, Cambridge, MA,1995.
%
% Example:
% M = [1 2 3 1]'; % 3D point (X=1,Y=2,Z=3)
% P1 = [rand(2,4);0 0 0 1]; % proyection matrix for view 1
% P2 = [rand(2,4);0 0 0 1]; % proyection matrix for view 2
% P3 = [rand(2,4);0 0 0 1]; % proyection matrix for view 3
% P4 = [rand(2,4);0 0 0 1]; % proyection matrix for view 4
% m1 = P1*M; % proyection point in view 1
% m2 = P2*M; % proyection point in view 2
% m3 = P3*M; % proyection point in view 3
% m4 = P4*M; % proyection point in view 4
% P = [P1;P2;P3;P4]; % all projection matrices
% m = [m1 m2 m3 m4]; % all proyection points
% [Ms,err,ms] = Bmv_reco3dn(m,P) % 3D reconstruction
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [M,err,ms] = Bmv_reco3dna(m,P)
n = size(m,2); % number of images
Q = zeros(2*n,3);
r = zeros(2*n,1);
for k = 1:n
p = P(k*3-2:k*3,:);
Q(k*2-1:k*2,:) = -p(1:2,1:3);
r(k*2-1:k*2,:) = p(1:2,4) - m(1:2,k);
end
M = [(Q'*Q)\Q'*r; 1];
ms = zeros(3,n);
ms(:) = P*M;
d = ms(1:2,:)-m(1:2,:);
err = sqrt(sum(d.*d,1));
|
github
|
domingomery/Balu-master
|
Bmv_tqsift.m
|
.m
|
Balu-master/MultiView/Bmv_tqsift.m
| 4,170 |
utf_8
|
1849d73ac546ca5daf5e04b4877f1f29
|
% D = Bmv_tqsift(Iq,It,options)
%
% Toolbox: Balu
%
% Search of image query (Iq) in target image (It) using SIFT.
%
% options.q : sliding windows's size in pixels
% options.d : sliding step in pixels
% options.nkp : minimal number of matching keypoints
% options.fast : '1' computes all SIFT keypoints of It at once
% '0' computes the SIFT keypoints for each sliding window
% options.show : display results
% options.roi : region of interest where the matching in target image
% will be searched. If roi is not given, it will be
% considered that the search are is the whole image It.
%
% D is the detection map.
%
%
% Example 1:
% I = imread('X1.png');
% Iq = I(165:194,80:109);
% It = imread('X2.png');
% op.q=30; op.d=5; op.show=1;op.nkp=2;op.fast=1;
% D = Bmv_tqsift(Iq,It,op);
% figure
% Bio_edgeview(It,bwperim(D>18))
%
% Example 2:
% I = imread('X1.png');
% ix = 315:394; jx=60:139;
% m1 = [mean(ix) mean(jx) 1]';
% Iq = I(ix,jx);
% It = imread('X2.png');
% figure(1);
% imshow(I,[]);
% title('Query image: blue box')
% hold on;
% plot(m1(2),m1(1),'rx')
% plot([min(jx) min(jx) max(jx) max(jx) min(jx)],[max(ix) min(ix) min(ix) max(ix) max(ix)])
% figure(2);
% imshow(It,[]);
% title('Target image')
% enterpause
% disp('Searching matching region without epiolar restriction...')
% op.q=80; op.d=5; op.show=1;op.nkp=3;op.fast=0;
% D1 = Bmv_tqsift(Iq,It,op);
% figure(3)
% Bio_edgeview(It,bwperim(D1>10))
% title('without epiplar line')
% enterpause
% disp('Searching matching region with epiolar restriction...')
% close all
% F = Bmv_fundamentalSIFT(I,It); % F matrix estimation
% ell = F*m1; % epipolar line
% R = Bmv_line2img(ell,size(It));
% op.roi = imdilate(R,ones(20,20));
% D2 = Bmv_tqsift(Iq,It,op);
% figure(3)
% Bio_edgeview(It,bwperim(D2>10))
% title('with epiplar line')
% hold on
% Bmv_epiplot(F,m1);
%
% (c) D.Mery, PUC-DCC, 2011
% http://dmery.ing.puc.cl
function D = Bmv_tqsift(Iq,It,options)
q = options.q; % sliding windows's size
d = options.d; % sliding step
nm = options.nkp; % minimal number of matching keypoints
show = options.show;
[N,M] = size(It);
D = zeros(N,M);
if ~isfield(options,'roi')
Iroi = ones(N,M);
else
Iroi = options.roi;
end
q2 = fix(q/2);
if show
figure(1)
clf
imshow(Iq,[])
title('Query image');
hold on
figure(2)
clf
imshow(It,[])
title('Target image');
hold on
drawnow
end
if size(It,3)==3
It = single(rgb2gray(It));
else
It = single(It);
end
if size(Iq,3)==3
Iq = single(rgb2gray(Iq)) ;
else
Iq = single(Iq);
end
if show
disp('Bmv_tqsift: computing SIFT descriptors...');
end
[fq, dq] = vl_sift(Iq) ;
if ~options.fast
for i=1:d:N-q+1
for j=1:d:M-q+1
if Iroi(i+q2,j+q2)
Itij = It(i:i+q-1,j:j+q-1);
[ftij, dtij] = vl_sift(Itij);
matches = vl_ubcmatch(dtij,dq);
if length(matches)>nm
D(i:i+q-1,j:j+q-1)=D(i:i+q-1,j:j+q-1)+1;
if show
plot(j+q2,i+q2,'go');
drawnow
end
end
end
end
end
else
[ft, dt] = vl_sift(It) ;
jt = ft(1,:);
it = ft(2,:);
for i=1:d:N-q+1
ii = find(and(it>=i,it<i+q));
dti = dt(:,ii);
jti = jt(ii);
for j=1:d:M-q+1
if Iroi(i+q2,j+q2)
dtij = dti(:,and(jti>=j,jti<j+q));
matches = vl_ubcmatch(dtij,dq);
if length(matches)>nm
D(i:i+q-1,j:j+q-1)=D(i:i+q-1,j:j+q-1)+1;
if show
plot(j+q2,i+q2,'go');
drawnow
end
end
end
end
end
end
|
github
|
domingomery/Balu-master
|
Bmv_projective2D.m
|
.m
|
Balu-master/MultiView/Bmv_projective2D.m
| 2,069 |
utf_8
|
2d4ea0df539208a583752e87f6bec971
|
% J = Bmv_projective2D(I,H,SJ,show)
%
% Toolbox: Balu
%
% 2D proyective transformation.
%
% J = projective2D(I,H,SJ,show) returns a new image J that is computed
% from the 2D projective transformation H of I.
%
% SJ is [NJ MJ] the size of the transformed image J. The
% default of SJ is [NJ,MJ] = size(I).
%
% show = 1 displays images I and J.
%
% The coordinates of I are (xp,yp) (from x',y'), and the
% coordinates of J are (x,y). The relation between both
% coordinate systems is: mp = H*m, where mp = [xp yp 1]'
% and m = [x y 1]'.
%
% R. Hartley and A. Zisserman. Multiple View Geometry in Computer
% Vision. Cambridge University Press, 2000.
%
% Example:
% I = rgb2gray(imread('testimg4.jpg')); % original image
% s = 7.5; t = pi/4; % scale and orientation
% H = [s*cos(t) -s*sin(t) -500
% s*sin(t) s*cos(t) -1750
% 0 0 1]; % similarity matrix
% J = Bmv_projective2D(I,H,[600 250],1); % projective transformation
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function [J,R] = Bmv_projective2D(I,H,SJ,show)
I = double(I);
[NI,MI] = size(I);
if ~exist('SJ','var')
[NJ,MJ] = size(I);
else
NJ = SJ(1);
MJ = SJ(2);
end
if ~exist('show','var')
show = 0;
end
if (show)
figure(1)
imshow(I,[])
title('image original')
axis on
end
J = mean(I(:))*ones(NJ,MJ);
R = zeros(NJ,MJ);
X = (1:NJ)'*ones(1,MJ);
Y = ones(NJ,1)*(1:MJ);
x = X(:);
y = Y(:);
m = [x'; y'; ones(1,NJ*MJ)];
mp = H*m;
mp = mp./(ones(3,1)*mp(3,:));
mp = fix(mp + [0.5 0.5 0]'*ones(1,NJ*MJ));
mm = [mp(1:2,:); x'; y'];
mm = mm(:,mm(1,:)>0);
mm = mm(:,mm(2,:)>0);
mm = mm(:,mm(1,:)<=NI);
mm = mm(:,mm(2,:)<=MI);
xp = mm(1,:);
yp = mm(2,:);
x = mm(3,:);
y = mm(4,:);
i = xp + (yp-1)*NI;
j = x + (y-1)*NJ;
J(j) = I(i);
R(j) = 1;
if (show)
figure(2)
imshow(J,[])
axis on
title('transformed image')
end
|
github
|
domingomery/Balu-master
|
Bmv_guihomography.m
|
.m
|
Balu-master/MultiView/Bmv_guihomography.m
| 12,514 |
utf_8
|
e3256a2abe57f425b28ed28cb4c4ba1d
|
function varargout = Bmv_guihomography(varargin)
% BMV_GUIHOMOGRAPHY M-file for Bmv_guihomography.fig
% BMV_GUIHOMOGRAPHY, by itself, creates a new BMV_GUIHOMOGRAPHY or raises the existing
% singleton*.
%
% H = BMV_GUIHOMOGRAPHY returns the handle to a new BMV_GUIHOMOGRAPHY or the handle to
% the existing singleton*.
%
% BMV_GUIHOMOGRAPHY('CALLBACK',hObject,eventData,handles,...) calls the local
% function named CALLBACK in BMV_GUIHOMOGRAPHY.M with the given input arguments.
%
% BMV_GUIHOMOGRAPHY('Property','Value',...) creates a new BMV_GUIHOMOGRAPHY or raises the
% existing singleton*. Starting from the left, property value pairs are
% applied to the GUI before matrixH_OpeningFunction gets called. An
% unrecognized property name or invalid value makes property application
% stop. All inputs are passed to Bmv_guihomography_OpeningFcn via varargin.
%
% *See GUI Options on GUIDE's Tools menu. Choose "GUI allows only one
% instance to run (singleton)".
%
% See also: GUIDE, GUIDATA, GUIHANDLES
% Copyright 2002-2003 The MathWorks, Inc.
% Edit the above text to modify the response to help Bmv_guihomography
% Last Modified by GUIDE v2.5 10-Nov-2014 09:03:20
% Begin initialization code - DO NOT EDIT
gui_Singleton = 1;
gui_State = struct('gui_Name', mfilename, ...
'gui_Singleton', gui_Singleton, ...
'gui_OpeningFcn', @Bmv_guihomography_OpeningFcn, ...
'gui_OutputFcn', @Bmv_guihomography_OutputFcn, ...
'gui_LayoutFcn', [] , ...
'gui_Callback', []);
if nargin && ischar(varargin{1})
gui_State.gui_Callback = str2func(varargin{1});
end
if nargout
[varargout{1:nargout}] = gui_mainfcn(gui_State, varargin{:});
else
gui_mainfcn(gui_State, varargin{:});
end
% End initialization code - DO NOT EDIT
% --- Executes just before Bmv_guihomography is made visible.
function Bmv_guihomography_OpeningFcn(hObject, eventdata, handles, varargin)
% This function has no output args, see OutputFcn.
% hObject handle to figure
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% varargin command line arguments to Bmv_guihomography (see VARARGIN)
% Choose default command line output for Bmv_guihomography
handles.output = hObject;
% Update handles structure
guidata(hObject, handles);
% UIWAIT makes Bmv_guihomography wait for user response (see UIRESUME)
% uiwait(handles.figure1);
dx = 0;
dy = 0;
th = 0;
u0 = 0;
v0 = 0;
z = 1;
x_p = [92 389 420 14]';
y_p = [57 71 305 278]';
save points dx dy th u0 v0 z x_p y_p
% --- Outputs from this function are returned to the command line.
function varargout = Bmv_guihomography_OutputFcn(hObject, eventdata, handles)
% varargout cell array for returning output args (see VARARGOUT);
% hObject handle to figure
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
% Get default command line output from handles structure
varargout{1} = handles.output;
% --- Executes on button press in left.
function left_Callback(hObject, eventdata, handles)
% hObject handle to left (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
dx = dx-[0 0 -20 20]';
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in right.
function right_Callback(hObject, eventdata, handles)
% hObject handle to right (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
dx = dx+[0 0 -20 20]';
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in A.
function A_Callback(hObject, eventdata, handles)
% hObject handle to A (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
dy = dy-[-20 0 0 20]';
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in V.
function V_Callback(hObject, eventdata, handles)
% hObject handle to V (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
dy = dy-[20 0 0 -20]';
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in points.
function points_Callback(hObject, eventdata, handles)
% hObject handle to points (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
disp('Select 4 points this way and press <Enter>:')
disp(' ');
disp(' 1------------4');
disp(' | |');
disp(' | |');
disp(' | |');
disp(' 2------------3');
figure(1)
[y_p,x_p] = getpts;
dy = 0;
dx = 0;
th = 0;
z = 1;
u0 = 0;
v0 = 0;
save points u0 v0 z th dx dy x_p y_p
% --- Executes on button press in Load.
function Load_Callback(hObject, eventdata, handles)
% hObject handle to Load (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
img_name = input('File name: ');
I = imread(img_name);
[N,M,P] = size(I);
if P==3
I = rgb2gray(I);
end
if (N>500)
I = imresize(I,500/N);
end
figure(1)
imshow(I)
title('original image')
save I I
% --- Executes on button press in rotplus.
function rotplus_Callback(hObject, eventdata, handles)
% hObject handle to rotplus (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
th = th+pi/40;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
function Hc = matrixHxy(x,y,x_p,y_p)
m1 = [x(1) x(2) x(3) x(4)
y(1) y(2) y(3) y(4)
1 1 1 1 ];
m2 = [x_p(1) x_p(2) x_p(3) x_p(4)
y_p(1) y_p(2) y_p(3) y_p(4)
1 1 1 1 ];
Hc = Bmv_homographySVD(m1,m2);
function hshape(x,y)
load I
load points
[N,M] = size(I);
% A1=[x(1) y(1) 1 0 0 0 -x(1)*x_p(1) -y(1)*x_p(1);
% 0 0 0 x(1) y(1) 1 -x(1)*y_p(1) -y(1)*y_p(1)];
% A2=[x(2) y(2) 1 0 0 0 -x(2)*x_p(2) -y(2)*x_p(2);
% 0 0 0 x(2) y(2) 1 -x(2)*y_p(2) -y(2)*y_p(2)];
% A3=[x(3) y(3) 1 0 0 0 -x(3)*x_p(3) -y(3)*x_p(3);
% 0 0 0 x(3) y(3) 1 -x(3)*y_p(3) -y(3)*y_p(3)];
% A4=[x(4) y(4) 1 0 0 0 -x(4)*x_p(4) -y(4)*x_p(4);
% 0 0 0 x(4) y(4) 1 -x(4)*y_p(4) -y(4)*y_p(4)];
% b=[x_p(1) y_p(1) x_p(2) y_p(2) x_p(3) y_p(3) x_p(4) y_p(4)];
%
% h=inv([A1;A2;A3;A4])*b';
%
% Hc =[h(1) h(2) h(3); h(4) h(5) h(6);h(7) h(8) 1];
% disp('aqui')
% m1 = [x(1) x(2) x(3) x(4)
% y(1) y(2) y(3) y(4)
% 1 1 1 1 ];
%
% m2 = [x_p(1) x_p(2) x_p(3) x_p(4)
% y_p(1) y_p(2) y_p(3) y_p(4)
% 1 1 1 1 ];
Hc = matrixHxy(x,y,x_p,y_p);
%Hc = Bmv_homographySVD(m1,m2);
Ht = [1 0 -N/2;0 1 -M/2;0 0 1];
Hr = [cos(th) -sin(th) 0
sin(th) cos(th) 0
0 0 1];
Hz = [z 0 u0;0 z v0; 0 0 1];
HH = Hz*Hc*inv(Ht)*Hr*Ht;
[N,M] = size(I);
J = Bmv_projective2D(I,HH,[N M],0);
save J J
figure(2)
imshow(uint8(J))
title('transformed image')
% --- Executes on button press in rotminus.
function rotminus_Callback(hObject, eventdata, handles)
% hObject handle to rotminus (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
th = th-pi/40;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in Save.
function Save_Callback(hObject, eventdata, handles)
% hObject handle to Save (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
img_format = input('Image format: ');
img_name = input('Image name : ');
load J
imwrite([img_name '.' img_format],img_format);
% --- Executes on button press in zoomplus.
function zoomplus_Callback(hObject, eventdata, handles)
% hObject handle to zoomplus (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
z = z/1.1;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in zoomminus.
function zoomminus_Callback(hObject, eventdata, handles)
% hObject handle to zoomminus (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
z = z*1.1;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in shiftleft.
function shiftleft_Callback(hObject, eventdata, handles)
% hObject handle to shiftleft (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
v0 = v0+50*z;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in shiftdown.
function shiftdown_Callback(hObject, eventdata, handles)
% hObject handle to shiftdown (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
u0 = u0-50*z;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in shiftright.
function shiftright_Callback(hObject, eventdata, handles)
% hObject handle to shiftright (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
v0 = v0-50*z;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in shiftup.
function shiftup_Callback(hObject, eventdata, handles)
% hObject handle to shiftup (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
u0 = u0+50*z;
x = x_p+dx;
y = y_p+dy;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
% --- Executes on button press in Ideal.
function Ideal_Callback(hObject, eventdata, handles)
% hObject handle to Ideal (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load I
load points
[N,M] = size(I);
x = [1 N N 1]';
y = [1 1 M M]';
% A1=[x(1) y(1) 1 0 0 0 -x(1)*x_p(1) -y(1)*x_p(1);
% 0 0 0 x(1) y(1) 1 -x(1)*y_p(1) -y(1)*y_p(1)];
% A2=[x(2) y(2) 1 0 0 0 -x(2)*x_p(2) -y(2)*x_p(2);
% 0 0 0 x(2) y(2) 1 -x(2)*y_p(2) -y(2)*y_p(2)];
% A3=[x(3) y(3) 1 0 0 0 -x(3)*x_p(3) -y(3)*x_p(3);
% 0 0 0 x(3) y(3) 1 -x(3)*y_p(3) -y(3)*y_p(3)];
% A4=[x(4) y(4) 1 0 0 0 -x(4)*x_p(4) -y(4)*x_p(4);
% 0 0 0 x(4) y(4) 1 -x(4)*y_p(4) -y(4)*y_p(4)];
% b=[x_p(1) y_p(1) x_p(2) y_p(2) x_p(3) y_p(3) x_p(4) y_p(4)];
%
% h=inv([A1;A2;A3;A4])*b';
%
% Hc =[h(1) h(2) h(3); h(4) h(5) h(6);h(7) h(8) 1];
Hc = matrixHxy(x,y,x_p,y_p);
% [N,M] = size(I);
J = Bmv_projective2D(I,Hc,[N M],0);
save J J
figure(3)
imshow(uint8(J),[])
title('transformed image (ideal)')
% --- Executes on button press in Restore.
function Restore_Callback(hObject, eventdata, handles)
% hObject handle to Restore (see GCBO)
% eventdata reserved - to be defined in a future version of MATLAB
% handles structure with handles and user data (see GUIDATA)
load points
dy = 0;
dx = 0;
th = 0;
z = 1;
u0 = 0;
v0 = 0;
x = x_p;
y = y_p;
save points u0 v0 z th dx dy x_p y_p
hshape(x,y)
|
github
|
domingomery/Balu-master
|
Bmv_antisimetric.m
|
.m
|
Balu-master/MultiView/Bmv_antisimetric.m
| 654 |
utf_8
|
86236235c55d1fd69c735d673022fa1c
|
% U = Bmv_antisimetric(u)
%
% Toolbox: Balu
%
% Antisimetric matrix
%
% antisimetric(u) returns the antisimetric matrix of a
% a 3x1 vector u.
%
% U = antisimetric(u) is a 3x3 matrix that
% U*v = cross(u,v) where cross(u,v) is the cross
% product between u and a 3x1 vector v.
%
% Example:
% u = [1 2 3]'
% v = rand(3,1)
% w1 = cross(u,v)
% U = Bmv_antisimetric(u)
% w2 = U*v % w1 must be equal to w2
%
% (c) D.Mery, PUC-DCC, 2010
% http://dmery.ing.puc.cl
function U = Bmv_antisimetric(u)
U = [ 0 -u(3) u(2)
u(3) 0 -u(1)
-u(2) u(1) 0 ];
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.