ocr_engine.py

import easyocr
import numpy as np
import cv2
import re

reader = easyocr.Reader(['en'], gpu=False)

def extract_weight_from_image(pil_img):
    try:
        # Convert PIL to NumPy
        img = np.array(pil_img)

        # Step 1: Preprocessing
        img = cv2.resize(img, None, fx=3.5, fy=3.5, interpolation=cv2.INTER_LINEAR)
        gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

        # Improve contrast & threshold
        blur = cv2.GaussianBlur(gray, (5, 5), 0)
        _, binary = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
        binary = cv2.bitwise_not(binary)

        # Step 2: OCR with bounding boxes
        results = reader.readtext(binary, detail=1)

        # Step 3: Filter for weight-like values
        weight_candidates = []
        for bbox, text, conf in results:
            clean = text.lower().replace("kg", "").replace("kgs", "").strip()
            clean = clean.replace("o", "0").replace("O", "0")  # common OCR mistake

            # Match like 2 digits or 3 digits or decimal numbers
            if re.fullmatch(r"\d{2,4}(\.\d{1,2})?", clean):
                weight_candidates.append((clean, conf))

        if not weight_candidates:
            return "Not detected", 0.0

        # Step 4: Pick most confident
        best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]
        return best_weight, round(best_conf * 100, 2)

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