Fix S/$ OCR confusion: add _fix_ocr_text method to convert S154.06 to $154.06

app.py

@@ -742,7 +742,14 @@ class ReceiptOCR:
     
     def postprocess_receipt(self, ocr_results):
         """Extract structured fields from OCR results with improved patterns."""
-        full_text = ' '.join([r['text'] for r in ocr_results])
+        # Fix common OCR errors (S->$ in amounts)
+        fixed_results = []
+        for r in ocr_results:
+            fixed_r = r.copy()
+            fixed_r['text'] = self._fix_ocr_text(r['text'])
+            fixed_results.append(fixed_r)
+        
+        full_text = ' '.join([r['text'] for r in fixed_results])
         
         fields = {
             'vendor': self._extract_vendor(ocr_results),
@@ -785,28 +792,36 @@ class ReceiptOCR:
         return None
     
     def _extract_total(self, text):
-        """Extract total amount - improved to find largest amount near TOTAL keyword."""
-        # First, find all dollar amounts in the text
-        all_amounts = re.findall(r'\$(\d{1,3}(?:,\d{3})*(?:\.\d{2})?)', text)
-        all_amounts = [float(a.replace(',', '')) for a in all_amounts]
+        """Extract total amount - handles S/$ OCR confusion."""
+        # Fix S -> $ in amounts (common OCR error)
+        fixed_text = re.sub(r'\bS(\d{1,3}(?:,\d{3})*(?:\.\d{2})?)\b', r'$\1', text)
+        
+        # Find all dollar amounts (now with fixed $ symbols)
+        all_amounts = re.findall(r'[\$S](\d{1,3}(?:,\d{3})*(?:\.\d{2})?)', fixed_text)
+        all_amounts = [float(a.replace(',', '')) for a in all_amounts if a]
+        
+        if not all_amounts:
+            # Try finding any decimal amounts
+            all_amounts = re.findall(r'(\d{1,3}(?:,\d{3})*\.\d{2})', fixed_text)
+            all_amounts = [float(a.replace(',', '')) for a in all_amounts if a]
         
         if not all_amounts:
             return None
         
         # Look for "TOTAL", "AMOUNT DUE", "BALANCE" keywords and find amount near them
-        lines = text.split('\n')
+        lines = fixed_text.split('\n')
         for i, line in enumerate(lines):
             line_upper = line.upper()
             if any(keyword in line_upper for keyword in ['TOTAL', 'AMOUNT DUE', 'BALANCE DUE', 'DUE']):
                 # Check this line and next 2 lines for amount
                 search_text = ' '.join(lines[i:min(i+3, len(lines))])
-                matches = re.findall(r'\$(\d{1,3}(?:,\d{3})*(?:\.\d{2})?)', search_text)
+                # Match both $ and S followed by amounts
+                matches = re.findall(r'[\$S](\d{1,3}(?:,\d{3})*(?:\.\d{2})?)', search_text)
                 if matches:
                     amounts_near_total = [float(m.replace(',', '')) for m in matches]
-                    # Return largest amount near TOTAL keyword
                     return f"{max(amounts_near_total):.2f}"
         
-        # Fallback: return largest amount overall (usually the total)
+        # Fallback: return largest amount overall
         return f"{max(all_amounts):.2f}"
     
     def _extract_time(self, text):
@@ -820,6 +835,14 @@ class ReceiptOCR:
             if match:
                 return match.group(0)
         return None
+    
+    def _fix_ocr_text(self, text):
+        """Fix common OCR errors like S->$ in amounts."""
+        # Fix S followed by digits -> $ (e.g., S154.06 -> $154.06)
+        text = re.sub(r'\bS(\d{1,3}(?:,\d{3})*(?:\.\d{2})?)\b', r'$\1', text)
+        # Fix Subtolal -> Subtotal (common OCR error)
+        text = re.sub(r'\bSubtolal\b', 'Subtotal', text, flags=re.IGNORECASE)
+        return text
 
 class LayoutLMFieldExtractor:
     """LayoutLMv3-based field extractor using fine-tuned weights if available."""