Update ocr_engine.py

ocr_engine.py

@@ -9,31 +9,33 @@ def extract_weight_from_image(pil_img):
     try:
         img = np.array(pil_img)
 
-        # Convert to grayscale and resize for better clarity
+        # Convert to grayscale and resize
         gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
         gray = cv2.resize(gray, None, fx=2, fy=2, interpolation=cv2.INTER_CUBIC)
 
-        # Histogram equalization
+        # Histogram equalization and adaptive threshold
         gray = cv2.equalizeHist(gray)
-
-        # Adaptive threshold to enhance text
         thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
                                        cv2.THRESH_BINARY, 11, 2)
-
-        # Invert for LCD-like contrast
         thresh = cv2.bitwise_not(thresh)
 
-        # OCR read
-        result = reader.readtext(thresh, detail=0)
-        combined_text = " ".join(result)
-        print("OCR Text:", combined_text)
-
-        # Regex to match weight like 25, 46.5, 75.45 etc.
-        match = re.search(r"\b\d{2,4}\.?\d{0,2}\b", combined_text)
-        if match:
-            return match.group(), 95.0
-        else:
-            return "No weight detected", 0.0
+        # OCR with bounding boxes
+        results = reader.readtext(thresh)
+
+        # Filter potential weight values
+        candidates = []
+        for (bbox, text, confidence) in results:
+            # Clean text
+            clean_text = text.replace('kg', '').strip()
+            if re.fullmatch(r"\d{2,4}(\.\d{1,2})?", clean_text):
+                candidates.append((clean_text, confidence))
+
+        if not candidates:
+            return "Not detected", 0.0
+
+        # Choose the highest confidence match
+        best_weight, conf = sorted(candidates, key=lambda x: -x[1])[0]
+        return best_weight, round(conf, 2)
 
     except Exception as e:
         return f"Error: {str(e)}", 0.0