ocr_engine.py

import easyocr
import numpy as np
import cv2
import re

# Load OCR engine
reader = easyocr.Reader(['en'], gpu=False)

def extract_weight_from_image(pil_img):
    try:
        img = np.array(pil_img)

        # Resize large image if needed
        max_dim = 1000
        height, width = img.shape[:2]
        if max(height, width) > max_dim:
            scale = max_dim / max(height, width)
            img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_AREA)

        # OCR without heavy preprocessing
        results = reader.readtext(img)
        print("DEBUG OCR RESULTS:", results)

        raw_texts = []
        weight_candidates = []

        for _, text, conf in results:
            original = text
            cleaned = text.lower().strip()

            # Fix common OCR mistakes
            cleaned = cleaned.replace(",", ".")
            cleaned = cleaned.replace("o", "0").replace("O", "0")
            cleaned = cleaned.replace("s", "5").replace("S", "5")
            cleaned = cleaned.replace("g", "9").replace("G", "6")
            cleaned = cleaned.replace("kg", "").replace("kgs", "")
            cleaned = re.sub(r"[^0-9\.]", "", cleaned)

            raw_texts.append(f"{original} → {cleaned} (conf: {round(conf, 2)})")

            # Match flexible weight formats like 75.02, 97.2, 102.34
            if cleaned.count(".") <= 1 and re.match(r"^\d{2,4}(\.\d{1,3})?$", cleaned):
                weight_candidates.append((cleaned, conf))

        if not weight_candidates:
            return "Not detected", 0.0, "\n".join(raw_texts)

        # Get best weight
        best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]

        # Strip unnecessary leading zeros
        if "." in best_weight:
            int_part, dec_part = best_weight.split(".")
            int_part = int_part.lstrip("0") or "0"
            best_weight = f"{int_part}.{dec_part}"
        else:
            best_weight = best_weight.lstrip("0") or "0"

        return best_weight, round(best_conf * 100, 2), "\n".join(raw_texts)

    except Exception as e:
        return f"Error: {str(e)}", 0.0, "OCR failed"