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:
        img = np.array(pil_img)

        # Resize image for consistency
        if img.shape[1] > 1000:
            img = cv2.resize(img, (1000, int(img.shape[0] * 1000 / img.shape[1])))

        # Convert to grayscale
        gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

        # Upscale image
        gray = cv2.resize(gray, None, fx=4, fy=4, interpolation=cv2.INTER_LINEAR)

        # Histogram Equalization and slight blur
        gray = cv2.equalizeHist(gray)
        blurred = cv2.GaussianBlur(gray, (3, 3), 0)

        # Adaptive threshold
        thresh = cv2.adaptiveThreshold(blurred, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                                       cv2.THRESH_BINARY, 11, 2)

        # Invert if needed
        white_ratio = np.mean(thresh > 127)
        if white_ratio < 0.5:
            thresh = cv2.bitwise_not(thresh)

        # OCR
        result = reader.readtext(thresh, detail=0)
        print("🧠 OCR Raw Output:", result)
        combined_text = " ".join(result).strip()

        # Extract number
        match = re.search(r"(\d{1,4}(?:\.\d{1,2})?)", combined_text)
        if match:
            weight = match.group(1)
            return f"{weight} kg", 100.0
        else:
            return "No weight detected kg", 0.0

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