ocr_engine.py

import easyocr
import numpy as np
import cv2
import re
import logging

# Set up logging for debugging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

# Initialize EasyOCR
easyocr_reader = easyocr.Reader(['en'], gpu=False)

def estimate_brightness(img):
    """Estimate image brightness to detect illuminated displays"""
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    return np.mean(gray)

def detect_roi(img):
    """Detect and crop the region of interest (likely the digital display)"""
    try:
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        # Stricter threshold for bright areas
        brightness = estimate_brightness(img)
        thresh_value = 220 if brightness > 100 else 180
        _, thresh = cv2.threshold(gray, thresh_value, 255, cv2.THRESH_BINARY)
        # Morphological operations to connect digits
        kernel = np.ones((9, 9), np.uint8)
        dilated = cv2.dilate(thresh, kernel, iterations=3)
        # Find contours
        contours, _ = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
        if contours:
            # Filter contours by size and aspect ratio (typical for displays)
            valid_contours = [c for c in contours if cv2.contourArea(c) > 500]
            if valid_contours:
                for contour in sorted(valid_contours, key=cv2.contourArea, reverse=True):
                    x, y, w, h = cv2.boundingRect(contour)
                    aspect_ratio = w / h
                    if 1.5 <= aspect_ratio <= 4.0 and w > 50 and h > 30:  # Typical display aspect ratio
                        x, y = max(0, x-40), max(0, y-40)
                        w, h = min(w+80, img.shape[1]-x), min(h+80, img.shape[0]-y)
                        return img[y:y+h, x:x+w]
        return img
    except Exception as e:
        logging.error(f"ROI detection failed: {str(e)}")
        return img

def correct_seven_segment(text, bbox):
    """Correct common seven-segment misreads based on bounding box shape"""
    if "6" in text:
        # Check bounding box aspect ratio to differentiate "6" from "2"
        (x1, y1), (x2, y2), (x3, y3), (x4, y4) = bbox
        width = abs(x2 - x1)
        height = abs(y2 - y1)
        aspect_ratio = width / height if height > 0 else 1.0
        # "2" typically has a more rectangular shape in seven-segment
        if aspect_ratio > 0.5:  # Adjust based on typical "2" vs "6" shapes
            text = text.replace("6", "2")
    return text

def enhance_image(img, mode="standard"):
    """Enhance image with different modes for multi-scale processing"""
    try:
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

        if mode == "seven_segment":
            # Minimal preprocessing for seven-segment displays
            _, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
            return thresh
        elif mode == "minimal":
            # Very light preprocessing
            denoised = cv2.GaussianBlur(gray, (3, 3), 0)
            _, thresh = cv2.threshold(denoised, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
            return thresh
        elif mode == "high_contrast":
            denoised = cv2.bilateralFilter(gray, d=11, sigmaColor=100, sigmaSpace=100)
            clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8))
            thresh = clahe.apply(denoised)
        elif mode == "low_noise":
            denoised = cv2.bilateralFilter(gray, d=7, sigmaColor=50, sigmaSpace=50)
            clahe = cv2.createCLAHE(clipLimit=1.5, tileGridSize=(8, 8))
            thresh = clahe.apply(denoised)
        else:
            denoised = cv2.bilateralFilter(gray, d=9, sigmaColor=75, sigmaSpace=75)
            clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
            thresh = clahe.apply(denoised)

        if mode not in ["seven_segment", "minimal"]:
            thresh = cv2.adaptiveThreshold(thresh, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
                                          cv2.THRESH_BINARY, 11, 2)

        # Morphological operations
        kernel = np.ones((3, 3), np.uint8)
        morphed = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=1)

        # Skip sharpening for seven-segment and minimal modes
        if mode not in ["seven_segment", "minimal"]:
            brightness = estimate_brightness(img)
            sharpen_strength = 3 if brightness > 100 else 5
            sharpen_kernel = np.array([[0, -1, 0], [-1, sharpen_strength, -1], [0, -1, 0]])
            morphed = cv2.filter2D(morphed, -1, sharpen_kernel)

        # Dynamic resizing
        h, w = morphed.shape
        target_size = 800
        scale_factor = min(target_size / max(h, w), 2.0) if max(h, w) < 300 else min(target_size / max(h, w), 1.0)
        if scale_factor != 1.0:
            morphed = cv2.resize(morphed, None, fx=scale_factor, fy=scale_factor, 
                                interpolation=cv2.INTER_CUBIC if scale_factor > 1 else cv2.INTER_AREA)

        return morphed
    except Exception as e:
        logging.error(f"Image enhancement failed (mode={mode}): {str(e)}")
        return img

def extract_weight_from_image(pil_img):
    try:
        img = np.array(pil_img)
        img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

        # Estimate brightness for adaptive thresholding
        brightness = estimate_brightness(img)
        conf_threshold = 0.7 if brightness > 100 else 0.5  # Stricter for bright displays

        # Detect ROI
        roi_img = detect_roi(img)

        # Process multiple image versions
        images_to_process = [
            ("seven_segment", enhance_image(roi_img, mode="seven_segment"), {'contrast_ths': 0.2, 'adjust_contrast': 0.5, 'allowlist': '0123456789.'}),
            ("minimal", enhance_image(roi_img, mode="minimal"), {'contrast_ths': 0.2, 'adjust_contrast': 0.5, 'allowlist': '0123456789.'}),
            ("standard", enhance_image(roi_img, mode="standard"), {'contrast_ths': 0.1, 'adjust_contrast': 0.5}),
            ("high_contrast", enhance_image(roi_img, mode="high_contrast"), {'contrast_ths': 0.1, 'adjust_contrast': 0.5}),
            ("low_noise", enhance_image(roi_img, mode="low_noise"), {'contrast_ths': 0.1, 'adjust_contrast': 0.5}),
            ("original", roi_img, {'contrast_ths': 0.2, 'adjust_contrast': 0.5, 'allowlist': '0123456789.'})
        ]

        best_weight = None
        best_conf = 0.0
        best_score = 0.0

        for mode, proc_img, ocr_params in images_to_process:
            # EasyOCR detection
            results = easyocr_reader.readtext(proc_img, detail=1, paragraph=False, **ocr_params)
            
            for (bbox, text, conf) in results:
                # Apply seven-segment correction
                text = correct_seven_segment(text, bbox)
                original_text = text
                text = text.lower().strip()

                # Fix common OCR errors
                text = text.replace(",", ".").replace(";", ".")
                text = text.replace("o", "0").replace("O", "0")
                text = text.replace("s", "5").replace("S", "5")
                text = text.replace("g", "9").replace("G", "6")
                text = text.replace("l", "1").replace("I", "1")
                text = text.replace("b", "8").replace("B", "8")
                text = text.replace("z", "2").replace("Z", "2")
                text = text.replace("q", "9").replace("Q", "9")
                text = text.replace("kgs", "").replace("kg", "").replace("k", "")
                text = re.sub(r"[^\d\.]", "", text)

                # Regex for weight (0.0 to 9999.999)
                if re.fullmatch(r"\d{1,4}(\.\d{0,3})?", text):
                    try:
                        weight = float(text)
                        # Score based on realistic weight range (0.1–500 kg)
                        range_score = 1.0 if 0.1 <= weight <= 500 else 0.3
                        # Strongly prefer two-digit weights for scales
                        digit_score = 1.5 if 10 <= weight < 100 else 1.0
                        score = conf * range_score * digit_score
                        if score > best_score and conf > conf_threshold:
                            best_weight = text
                            best_conf = conf
                            best_score = score
                    except ValueError:
                        continue

        if not best_weight:
            logging.info("No valid weight detected")
            return "Not detected", 0.0

        # Format output
        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.rstrip('0')}"
        else:
            best_weight = best_weight.lstrip('0') or "0"

        return best_weight, round(best_conf * 100, 2)

    except Exception as e:
        logging.error(f"Weight extraction failed: {str(e)}")
        return "Not detected", 0.0