ocr_engine.py

import re
import cv2
import numpy as np
from PIL import Image
from paddleocr import PaddleOCR

# ✅ Initialize PaddleOCR once
ocr = PaddleOCR(use_angle_cls=False, lang='en', show_log=False)

def preprocess_image(image):
    """
    Convert to grayscale and enhance contrast to help OCR.
    """
    gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
    processed = cv2.adaptiveThreshold(
        gray, 255,
        cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
        cv2.THRESH_BINARY_INV,
        11, 2
    )
    return processed

def extract_weight_from_image(pil_image):
    try:
        # ✅ Convert to OpenCV format
        image = np.array(pil_image.convert("RGB"))

        # ✅ Preprocess image
        processed = preprocess_image(image)

        # Optional: Save debug image
        debug_path = "debug_processed_image.png"
        Image.fromarray(processed).save(debug_path)
        print(f"[DEBUG] Saved preprocessed image to {debug_path}")

        # ✅ Run OCR on original (RGB) image (not preprocessed)
        result = ocr.ocr(image, cls=False)

        print("🔍 PaddleOCR Results:")
        for line in result[0]:
            text = line[1][0]
            conf = line[1][1]
            print(f"  • Text: '{text}' | Confidence: {conf*100:.2f}%")

            # Try to extract number like 53.25 or 100
            match = re.search(r"\b\d{1,4}(\.\d{1,2})?\b", text)
            if match:
                return match.group(), round(conf * 100, 2)

        return "No weight detected", 0.0

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