Fix vendor extraction: detect known vendors like Einstein, skip numbers/IDs, prefer longer names. Remove Time field from display.

app.py

@@ -761,22 +761,59 @@ class ReceiptOCR:
         return fields
     
     def _extract_vendor(self, ocr_results):
-        """Extract vendor name, usually in first few lines."""
+        """Extract vendor name - look for business name in top portion of receipt."""
         if not ocr_results:
             return None
         
-        # Look for vendor in top 3 results (usually at top of receipt)
-        top_results = sorted(ocr_results, key=lambda x: x['bbox'][1])[:3]
+        # Sort by vertical position (top of receipt first)
+        sorted_results = sorted(ocr_results, key=lambda x: x['bbox'][1] if isinstance(x['bbox'], list) and len(x['bbox']) > 1 else 0)
         
+        # Look in top 10 results for vendor name
+        top_results = sorted_results[:10]
+        
+        # Skip words that are clearly not vendor names
+        skip_words = {'TOTAL', 'DATE', 'TIME', 'RECEIPT', 'THANK', 'YOU', 'STORE', 'HOST', 
+                      'ORDER', 'TYPE', 'TOGO', 'DINE', 'IN', 'CHECK', 'CLOSED', 'AMEX',
+                      'VISA', 'MASTERCARD', 'CASH', 'CHANGE', 'SUBTOTAL', 'TAX'}
+        
+        # Known vendor patterns (common stores)
+        known_vendors = ['EINSTEIN', 'STARBUCKS', 'MCDONALDS', 'WALMART', 'TARGET', 
+                        'CHIPOTLE', 'PANERA', 'DUNKIN', 'SUBWAY', 'CHICK-FIL-A']
+        
+        # First, check if any known vendor is in the OCR results
+        for result in top_results:
+            text = result['text'].strip().upper()
+            for vendor in known_vendors:
+                if vendor in text:
+                    return result['text'].strip()
+        
+        # Look for longest meaningful text (likely the business name)
+        candidates = []
         for result in top_results:
             text = result['text'].strip()
-            # Skip common non-vendor words
-            if text and len(text) > 2 and text.upper() not in ['TOTAL', 'DATE', 'TIME', 'RECEIPT', 'THANK', 'YOU']:
-                # Take longest text as vendor (usually company name)
-                if len(text) > 5:
-                    return text
+            text_upper = text.upper()
+            
+            # Skip short texts, numbers, and common skip words
+            if len(text) < 3:
+                continue
+            if text_upper in skip_words:
+                continue
+            if re.match(r'^[\d\s\-\/\.\$\,]+$', text):  # Skip pure numbers/symbols
+                continue
+            if re.match(r'^#?\d+$', text):  # Skip store numbers like #2846
+                continue
+                
+            # Prefer texts with letters and reasonable length
+            if len(text) >= 4 and any(c.isalpha() for c in text):
+                candidates.append((text, len(text), result['confidence']))
+        
+        # Return the longest candidate with good confidence
+        if candidates:
+            # Sort by length (longer = more likely to be full vendor name)
+            candidates.sort(key=lambda x: (x[1], x[2]), reverse=True)
+            return candidates[0][0]
         
-        return top_results[0]['text'] if top_results else None
+        return None
     
     def _extract_date(self, text):
         """Extract date with improved patterns."""
@@ -1298,7 +1335,6 @@ def process_receipt(image):
             ('Vendor', fields.get('vendor')),
             ('Date', fields.get('date')),
             ('Total', f"${fields.get('total')}" if fields.get('total') else None),
-            ('Time', fields.get('time')),
         ]:
             display = value or '<span style="color: #9ca3af;">Not found</span>'
             fields_html += f"<div style='padding: 8px; background: #0f172a; color: #e5e7eb; border: 1px solid #1f2937; border-radius: 6px; margin: 4px 0;'><b>{name}:</b> {display}</div>"