Update ocr_engine.py

ocr_engine.py

@@ -31,31 +31,34 @@ def estimate_brightness(img):
     gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
     return np.mean(gray)
 
-def preprocess_image(img):
+def preprocess_image(img, scale=1.0):
     """Preprocess image for better OCR accuracy."""
+    if scale != 1.0:
+        img = cv2.resize(img, None, fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)
+        save_debug_image(img, f"01_preprocess_scaled_{scale}")
     gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
     # Apply bilateral filter to preserve edges
-    denoised = cv2.bilateralFilter(gray, 11, 17, 17)
-    save_debug_image(denoised, "01_preprocess_bilateral")
+    denoised = cv2.bilateralFilter(gray, 9, 15, 15)
+    save_debug_image(denoised, "02_preprocess_bilateral")
     # Enhance contrast using CLAHE
-    clahe = cv2.createCLAHE(clipLimit=2.5, tileGridSize=(8, 8))
+    clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8, 8))
     enhanced = clahe.apply(denoised)
-    save_debug_image(enhanced, "02_preprocess_clahe")
+    save_debug_image(enhanced, "03_preprocess_clahe")
     # Sharpen the image
     kernel_sharpening = np.array([[-1, -1, -1], [-1, 9, -1], [-1, -1, -1]])
     sharpened = cv2.filter2D(enhanced, -1, kernel_sharpening)
-    save_debug_image(sharpened, "03_preprocess_sharpened")
+    save_debug_image(sharpened, "04_preprocess_sharpened")
     return sharpened
 
 def correct_rotation(img):
     """Correct image rotation using Hough Transform."""
     try:
-        edges = cv2.Canny(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), 100, 200)
-        lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold=100, minLineLength=100, maxLineGap=10)
+        edges = cv2.Canny(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), 50, 150)
+        lines = cv2.HoughLinesP(edges, 1, np.pi / 180, threshold=80, minLineLength=50, maxLineGap=10)
         if lines is not None:
             angles = [np.arctan2(line[0][3] - line[0][1], line[0][2] - line[0][0]) * 180 / np.pi for line in lines]
-            angle = np.median(angles)  # Use median for robustness
-            if abs(angle) > 5:
+            angle = np.median(angles)
+            if abs(angle) > 3:
                 (h, w) = img.shape[:2]
                 center = (w // 2, h // 2)
                 M = cv2.getRotationMatrix2D(center, angle, 1.0)
@@ -70,87 +73,96 @@ def correct_rotation(img):
 def detect_roi(img):
     """Detect and crop the region of interest (likely the digital display)."""
     try:
-        save_debug_image(img, "04_original")
-        preprocessed = preprocess_image(img)
+        save_debug_image(img, "05_original")
         brightness_map = cv2.GaussianBlur(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), (15, 15), 0)
         
-        # Dynamic adaptive thresholding
-        block_size = max(11, min(31, int(img.shape[0] / 20) * 2 + 1))
-        thresh = cv2.adaptiveThreshold(preprocessed, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
-                                       cv2.THRESH_BINARY_INV, block_size, 5)
-        _, otsu_thresh = cv2.threshold(preprocessed, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
-        combined_thresh = cv2.bitwise_and(thresh, otsu_thresh)
-        save_debug_image(combined_thresh, "05_roi_combined_threshold")
-        
-        # Morphological operations to connect digits
-        kernel = np.ones((5, 5), np.uint8)
-        dilated = cv2.dilate(combined_thresh, kernel, iterations=2)
-        eroded = cv2.erode(dilated, kernel, iterations=1)
-        save_debug_image(eroded, "06_roi_morphological")
-        
-        contours, _ = cv2.findContours(eroded, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
-        
-        if contours:
-            img_area = img.shape[0] * img.shape[1]
-            valid_contours = []
-            for c in contours:
-                area = cv2.contourArea(c)
-                x, y, w, h = cv2.boundingRect(c)
-                roi_brightness = np.mean(brightness_map[y:y+h, x:x+w])
-                aspect_ratio = w / h
-                if (500 < area < (img_area * 0.9) and 
-                    0.8 <= aspect_ratio <= 12.0 and w > 60 and h > 30 and roi_brightness > 80):
-                    valid_contours.append((c, roi_brightness))
-                    logging.debug(f"Contour: Area={area}, Aspect={aspect_ratio:.2f}, Brightness={roi_brightness:.2f}")
+        # Try multiple scales for preprocessing
+        scales = [1.0, 1.5, 0.75]
+        for scale in scales:
+            preprocessed = preprocess_image(img, scale)
+            block_size = max(11, min(31, int(img.shape[0] / 20) * 2 + 1))
+            thresh = cv2.adaptiveThreshold(preprocessed, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
+                                           cv2.THRESH_BINARY_INV, block_size, 5)
+            _, otsu_thresh = cv2.threshold(preprocessed, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
+            combined_thresh = cv2.bitwise_and(thresh, otsu_thresh)
+            save_debug_image(combined_thresh, f"06_roi_combined_threshold_scale_{scale}")
+            
+            # Morphological operations
+            kernel = np.ones((5, 5), np.uint8)
+            dilated = cv2.dilate(combined_thresh, kernel, iterations=2)
+            eroded = cv2.erode(dilated, kernel, iterations=1)
+            save_debug_image(eroded, f"07_roi_morphological_scale_{scale}")
+            
+            contours, _ = cv2.findContours(eroded, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
             
-            if valid_contours:
-                contour, _ = max(valid_contours, key=lambda x: x[1])  # Max brightness
-                x, y, w, h = cv2.boundingRect(contour)
-                padding = 100
-                x, y = max(0, x - padding), max(0, y - padding)
-                w, h = min(w + 2 * padding, img.shape[1] - x), min(h + 2 * padding, img.shape[0] - y)
-                roi_img = img[y:y+h, x:x+w]
-                save_debug_image(roi_img, "07_detected_roi")
-                logging.info(f"Detected ROI with dimensions: ({x}, {y}, {w}, {h})")
-                return roi_img, (x, y, w, h)
+            if contours:
+                img_area = img.shape[0] * img.shape[1]
+                valid_contours = []
+                for c in contours:
+                    area = cv2.contourArea(c)
+                    x, y, w, h = cv2.boundingRect(c)
+                    roi_brightness = np.mean(brightness_map[y:y+h, x:x+w] if scale == 1.0 else cv2.resize(brightness_map, (img.shape[1], img.shape[0])))
+                    aspect_ratio = w / h
+                    if (200 < area < (img_area * 0.95) and 
+                        0.5 <= aspect_ratio <= 15.0 and w > 50 and h > 20 and roi_brightness > 60):
+                        valid_contours.append((c, roi_brightness))
+                        logging.debug(f"Contour: Scale={scale}, Area={area}, Aspect={aspect_ratio:.2f}, Brightness={roi_brightness:.2f}")
+                
+                if valid_contours:
+                    contour, _ = max(valid_contours, key=lambda x: x[1])
+                    x, y, w, h = cv2.boundingRect(contour)
+                    if scale != 1.0:
+                        x, y, w, h = [int(v / scale) for v in (x, y, w, h)]
+                    padding = 120
+                    x, y = max(0, x - padding), max(0, y - padding)
+                    w, h = min(w + 2 * padding, img.shape[1] - x), min(h + 2 * padding, img.shape[0] - y)
+                    roi_img = img[y:y+h, x:x+w]
+                    save_debug_image(roi_img, f"08_detected_roi_scale_{scale}")
+                    logging.info(f"Detected ROI with dimensions: ({x}, {y}, {w}, {h}) at scale {scale}")
+                    return roi_img, (x, y, w, h)
         
         logging.info("No suitable ROI found, attempting fallback criteria.")
         # Fallback with relaxed criteria
-        valid_contours = [c for c in contours if 300 < cv2.contourArea(c) < (img_area * 0.95) and 
-                          0.5 <= cv2.boundingRect(c)[2]/cv2.boundingRect(c)[3] <= 15.0]
+        preprocessed = preprocess_image(img)
+        thresh = cv2.adaptiveThreshold(preprocessed, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
+                                       cv2.THRESH_BINARY_INV, block_size, 8)
+        save_debug_image(thresh, "06_roi_fallback_threshold")
+        contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+        valid_contours = [c for c in contours if 100 < cv2.contourArea(c) < (img.shape[0] * img.shape[1] * 0.95) and 
+                          0.3 <= cv2.boundingRect(c)[2]/cv2.boundingRect(c)[3] <= 20.0]
         if valid_contours:
             contour = max(valid_contours, key=cv2.contourArea)
             x, y, w, h = cv2.boundingRect(contour)
-            padding = 100
+            padding = 120
             x, y = max(0, x - padding), max(0, y - padding)
             w, h = min(w + 2 * padding, img.shape[1] - x), min(h + 2 * padding, img.shape[0] - y)
             roi_img = img[y:y+h, x:x+w]
-            save_debug_image(roi_img, "07_detected_roi_fallback")
+            save_debug_image(roi_img, "08_detected_roi_fallback")
             logging.info(f"Detected fallback ROI with dimensions: ({x}, {y}, {w}, {h})")
             return roi_img, (x, y, w, h)
         
         logging.info("No suitable ROI found, returning original image.")
-        save_debug_image(img, "07_no_roi_original_fallback")
+        save_debug_image(img, "08_no_roi_original_fallback")
         return img, None
     except Exception as e:
         logging.error(f"ROI detection failed: {str(e)}")
-        save_debug_image(img, "07_roi_detection_error_fallback")
+        save_debug_image(img, "08_roi_detection_error_fallback")
         return img, None
 
 def detect_segments(digit_img, brightness):
     """Detect seven-segment patterns in a digit image."""
     h, w = digit_img.shape
-    if h < 15 or w < 10:
+    if h < 10 or w < 8:
         return None
 
     segments = {
-        'top': (int(w*0.15), int(w*0.85), 0, int(h*0.2)),
-        'middle': (int(w*0.15), int(w*0.85), int(h*0.45), int(h*0.55)),
-        'bottom': (int(w*0.15), int(w*0.85), int(h*0.8), h),
-        'left_top': (0, int(w*0.25), int(h*0.15), int(h*0.5)),
-        'left_bottom': (0, int(w*0.25), int(h*0.5), int(h*0.85)),
-        'right_top': (int(w*0.75), w, int(h*0.15), int(h*0.5)),
-        'right_bottom': (int(w*0.75), w, int(h*0.5), int(h*0.85))
+        'top': (int(w*0.1), int(w*0.9), 0, int(h*0.25)),
+        'middle': (int(w*0.1), int(w*0.9), int(h*0.45), int(h*0.55)),
+        'bottom': (int(w*0.1), int看到的: int(w*0.9), int(h*0.75), h),
+        'left_top': (0, int(w*0.3), int(h*0.1), int(h*0.5)),
+        'left_bottom': (0, int(w*0.3), int(h*0.5), int(h*0.9)),
+        'right_top': (int(w*0.7), w, int(h*0.1), int(h*0.5)),
+        'right_bottom': (int(w*0.7), w, int(h*0.5), int(h*0.9))
     }
 
     segment_presence = {}
@@ -163,7 +175,7 @@ def detect_segments(digit_img, brightness):
             continue
         pixel_count = np.sum(region == 255)
         total_pixels = region.size
-        segment_presence[name] = pixel_count / total_pixels > (0.25 if brightness < 100 else 0.45)
+        segment_presence[name] = pixel_count / total_pixels > (0.2 if brightness < 80 else 0.4)
 
     digit_patterns = {
         '0': ('top', 'bottom', 'left_top', 'left_bottom', 'right_top', 'right_bottom'),
@@ -184,7 +196,7 @@ def detect_segments(digit_img, brightness):
         matches = sum(1 for segment in pattern if segment_presence.get(segment, False))
         non_matches_penalty = sum(1 for segment in segment_presence if segment not in pattern and segment_presence[segment])
         score = matches - 0.2 * non_matches_penalty
-        if matches >= len(pattern) * 0.75:
+        if matches >= len(pattern) * 0.7:
             score += 1.0
         if score > max_score:
             max_score = score
@@ -198,20 +210,20 @@ def custom_seven_segment_ocr(img, roi_bbox):
     try:
         preprocessed = preprocess_image(img)
         brightness = estimate_brightness(img)
-        thresh_value = 100 if brightness < 100 else 0
+        thresh_value = 80 if brightness < 80 else 0
         _, thresh = cv2.threshold(preprocessed, thresh_value, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
-        save_debug_image(thresh, "08_roi_thresh_for_digits")
+        save_debug_image(thresh, "09_roi_thresh_for_digits")
 
-        # Morphological operations to enhance digit segments
+        # Morphological operations
         kernel = np.ones((3, 3), np.uint8)
         thresh = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=2)
-        save_debug_image(thresh, "09_morph_closed")
+        save_debug_image(thresh, "10_morph_closed")
 
         batch_size = max(4, min(16, int(img.shape[0] * img.shape[1] / 100000)))
         results = easyocr_reader.readtext(thresh, detail=1, paragraph=False, 
-                                          contrast_ths=0.3, adjust_contrast=1.0, 
-                                          text_threshold=0.6, mag_ratio=3.0, 
-                                          allowlist='0123456789.', batch_size=batch_size, y_ths=0.2)
+                                          contrast_ths=0.2, adjust_contrast=1.2, 
+                                          text_threshold=0.5, mag_ratio=4.0, 
+                                          allowlist='0123456789.', batch_size=batch_size, y_ths=0.3)
         
         logging.info(f"EasyOCR results: {results}")
         if not results:
@@ -222,7 +234,7 @@ def custom_seven_segment_ocr(img, roi_bbox):
         for (bbox, text, conf) in results:
             (x1, y1), (x2, y2), (x3, y3), (x4, y4) = bbox
             h_bbox = max(y1, y2, y3, y4) - min(y1, y2, y3, y4)
-            if len(text) == 1 and (text.isdigit() or text == '.') and h_bbox > 8:
+            if (text.isdigit() or text == '.') and h_bbox > 6:
                 x_min, x_max = int(min(x1, x4)), int(max(x2, x3))
                 y_min, y_max = int(min(y1, y2)), int(max(y3, y4))
                 digits_info.append((x_min, x_max, y_min, y_max, text, conf))
@@ -235,8 +247,8 @@ def custom_seven_segment_ocr(img, roi_bbox):
             if x_max <= x_min or y_max <= y_min:
                 continue
             digit_img_crop = thresh[y_min:y_max, x_min:x_max]
-            save_debug_image(digit_img_crop, f"10_digit_crop_{idx}_{easyocr_char}")
-            if easyocr_conf > 0.95 or easyocr_char == '.':
+            save_debug_image(digit_img_crop, f"11_digit_crop_{idx}_{easyocr_char}")
+            if easyocr_conf > 0.9 or easyocr_char == '.':
                 recognized_text += easyocr_char
             else:
                 digit_from_segments = detect_segments(digit_img_crop, brightness)
@@ -268,18 +280,18 @@ def extract_weight_from_image(pil_img):
         img = correct_rotation(img)
 
         brightness = estimate_brightness(img)
-        conf_threshold = 0.7 if brightness > 150 else (0.6 if brightness > 80 else 0.4)
+        conf_threshold = 0.7 if brightness > 150 else (0.5 if brightness > 80 else 0.3)
 
         roi_img, roi_bbox = detect_roi(img)
         if roi_bbox:
             roi_area = roi_bbox[2] * roi_bbox[3]
-            conf_threshold *= 1.2 if roi_area > (img.shape[0] * img.shape[1] * 0.5) else 1.0
+            conf_threshold *= 1.1 if roi_area > (img.shape[0] * img.shape[1] * 0.5) else 1.0
 
         custom_result = custom_seven_segment_ocr(roi_img, roi_bbox)
         if custom_result:
             try:
                 weight = float(custom_result)
-                if 0.001 <= weight <= 1000:
+                if 0.001 <= weight <= 2000:
                     logging.info(f"Custom OCR result: {custom_result}, Confidence: 95.0%")
                     return custom_result, 95.0
                 else:
@@ -292,14 +304,24 @@ def extract_weight_from_image(pil_img):
         block_size = max(11, min(31, int(roi_img.shape[0] / 20) * 2 + 1))
         final_roi = cv2.adaptiveThreshold(preprocessed_roi, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
                                           cv2.THRESH_BINARY_INV, block_size, 8)
-        save_debug_image(final_roi, "11_fallback_adaptive_thresh")
+        save_debug_image(final_roi, "12_fallback_adaptive_thresh")
 
         batch_size = max(4, min(16, int(roi_img.shape[0] * roi_img.shape[1] / 100000)))
         results = easyocr_reader.readtext(final_roi, detail=1, paragraph=False, 
-                                          contrast_ths=0.4, adjust_contrast=1.2, 
-                                          text_threshold=0.5, mag_ratio=4.0, 
-                                          allowlist='0123456789. kglb', batch_size=batch_size, y_ths=0.2)
+                                          contrast_ths=0.3, adjust_contrast=1.2, 
+                                          text_threshold=0.4, mag_ratio=5.0, 
+                                          allowlist='0123456789. kglb', batch_size=batch_size, y_ths=0.3)
 
+        # Secondary EasyOCR pass with different parameters
+        if not results:
+            logging.info("First EasyOCR pass failed, trying with relaxed parameters.")
+            results = easyocr_reader.readtext(final_roi, detail=1, paragraph=False, 
+                                              contrast_ths=0.2, adjust_contrast=1.5, 
+                                              text_threshold=0.3, mag_ratio=6.0, 
+                                              allowlist='0123456789. kglb', batch_size=batch_size, y_ths=0.4)
+            save_debug_image(final_roi, "12_fallback_adaptive_thresh_relaxed")
+
+        logging.info(f"EasyOCR results: {results}")
         best_weight = None
         best_conf = 0.0
         best_score = 0.0
@@ -325,9 +347,9 @@ def extract_weight_from_image(pil_img):
                         weight /= 1000  # Convert grams to kilograms
                     elif unit == 'lb':
                         weight *= 0.453592  # Convert pounds to kilograms
-                    range_score = 1.5 if 0.001 <= weight <= 1000 else 0.8
+                    range_score = 1.5 if 0.001 <= weight <= 2000 else 0.7
                     digit_count = len(text.replace('.', ''))
-                    digit_score = 1.3 if 2 <= digit_count <= 7 else 0.9
+                    digit_score = 1.3 if 1 <= digit_count <= 8 else 0.8
                     score = conf * range_score * digit_score
                     if roi_bbox:
                         (x_roi, y_roi, w_roi, h_roi) = roi_bbox
@@ -335,8 +357,8 @@ def extract_weight_from_image(pil_img):
                         x_min, y_min = int(min(b[0] for b in bbox)), int(min(b[1] for b in bbox))
                         x_max, y_max = int(max(b[0] for b in bbox)), int(max(b[1] for b in bbox))
                         bbox_area = (x_max - x_min) * (y_max - y_min)
-                        if roi_area > 0 and bbox_area / roi_area < 0.05:
-                            score *= 0.6
+                        if roi_area > 0 and bbox_area / roi_area < 0.03:
+                            score *= 0.5
                     if score > best_score and conf > conf_threshold:
                         best_weight = text
                         best_conf = conf
@@ -361,8 +383,10 @@ def extract_weight_from_image(pil_img):
 
         try:
             final_weight = float(best_weight)
-            if final_weight < 0.001 or final_weight > 1000:
-                best_conf *= 0.7
+            if final_weight < 0.001 or final_weight > 2000:
+                best_conf *= 0.6
+            elif final_weight == 0 and best_conf < 0.9:
+                best_conf *= 0.7  # Penalize zero weights with low confidence
         except ValueError:
             pass