Update ocr_engine.py

ocr_engine.py

@@ -3,39 +3,42 @@ import numpy as np
 import cv2
 import re
 
-# Initialize OCR reader once
 reader = easyocr.Reader(['en'], gpu=False)
 
 def extract_weight_from_image(pil_img):
     try:
-        # Convert image to NumPy format
         img = np.array(pil_img)
 
-        # Resize and preprocess
+        # Resize and grayscale
         img = cv2.resize(img, None, fx=3.5, fy=3.5, interpolation=cv2.INTER_LINEAR)
         gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
+
+        # Denoise & Adaptive Threshold (better than OTSU for displays)
         gray = cv2.bilateralFilter(gray, 11, 17, 17)
-        _, thresh = cv2.threshold(gray, 120, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
+        thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
+                                       cv2.THRESH_BINARY_INV, 11, 2)
 
-        # OCR
+        # Run OCR
         results = reader.readtext(thresh)
 
-        # Debug
-        print("OCR Results:", results)
+        print("OCR Results:", results)  # For debugging
 
         weight_candidates = []
         for _, text, conf in results:
-            clean = text.lower().replace("kg", "").strip()
-            clean = clean.replace("o", "0").replace("O", "0")  # fix OCR misreads
+            cleaned = text.lower()
+            cleaned = cleaned.replace("kg", "").replace("kgs", "")
+            cleaned = cleaned.replace("o", "0").replace("O", "0")
+            cleaned = cleaned.replace("s", "5").replace("S", "5")
+            cleaned = cleaned.replace("g", "9").replace("G", "6")
+
+            cleaned = re.sub(r"[^\d\.]", "", cleaned)  # remove non-numeric
 
-            # Match weights like 86, 85.5, 102.3
-            if re.fullmatch(r"\d{2,4}(\.\d{1,2})?", clean):
-                weight_candidates.append((clean, conf))
+            if re.fullmatch(r"\d{2,4}(\.\d{1,2})?", cleaned):  # allow 2-4 digit weights
+                weight_candidates.append((cleaned, conf))
 
         if not weight_candidates:
             return "Not detected", 0.0
 
-        # Return best candidate
         best_weight, best_conf = sorted(weight_candidates, key=lambda x: -x[1])[0]
         return best_weight, round(best_conf * 100, 2)