Update ocr_engine.py

ocr_engine.py

@@ -3,42 +3,48 @@ from PIL import Image, ImageFilter
 import torch
 import re
 
-# Load TrOCR model
+# Load model
 processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten")
 model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten")
 
 def clean_ocr_text(text):
     print("[RAW OCR]", text)
-    # Fix common misreads
     text = text.replace(",", ".").replace("s", "5").replace("o", "0").replace("O", "0")
-    text = re.sub(r"[^\d\.kg]", "", text.lower())  # Keep digits, dot, kg
+    text = re.sub(r"[^\d.kg]", "", text.lower())  # Keep digits, dots, and kg
     print("[CLEANED OCR]", text)
     return text
 
-def preprocess_image(image):
-    # Enlarge + sharpen for better OCR
-    image = image.resize((image.width * 2, image.height * 2), Image.BICUBIC)
-    image = image.filter(ImageFilter.SHARPEN)
-    return image
+def restore_decimal(text):
+    if re.fullmatch(r"\d{5}", text):
+        return f"{text[:2]}.{text[2:]}"
+    elif re.fullmatch(r"\d{4}", text):
+        return f"{text[:2]}.{text[2:]}"
+    return text
 
 def extract_weight(image):
     try:
-        image = preprocess_image(image)
+        image = image.resize((image.width * 2, image.height * 2), Image.BICUBIC)
+        image = image.filter(ImageFilter.SHARPEN)
+
         pixel_values = processor(images=image, return_tensors="pt").pixel_values
         generated_ids = model.generate(pixel_values)
         raw_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
 
         cleaned = clean_ocr_text(raw_text)
 
-        # Try matching decimal weight with unit
-        match = re.search(r'(\d{1,3}\.\d{1,3})\s*(kg|g)', cleaned)
+        # Try direct match: e.g., 52.25 kg or 75.0 g
+        match = re.search(r"(\d{1,3}\.\d{1,3})\s*(kg|g)", cleaned)
         if match:
             return f"{match.group(1)} {match.group(2)}"
 
-        # Fallback: match only decimal number
-        fallback = re.search(r'(\d{1,3}\.\d{1,3})', cleaned)
-        if fallback:
-            return f"{fallback.group(1)} g"  # Default to grams
+        # Try fallback: extract digits and manually guess decimal
+        fallback_match = re.search(r"(\d{4,5})", cleaned)
+        if fallback_match:
+            fallback_value = restore_decimal(fallback_match.group(1))
+
+            # Check for presence of unit hints in raw_text
+            unit = "kg" if "kg" in raw_text.lower() else "g"
+            return f"{fallback_value} {unit}"
 
         return f"No valid weight found | OCR: {cleaned}"
     except Exception as e: