Added PanOCR refined code to do Pan Card ocr

.gitignore

@@ -1,2 +1,4 @@
 env
-.env
+.env
+*.pyc
+*.DS_Store

app/__init__.py

@@ -10,7 +10,7 @@ def create_app():
         # Load model once
         app.models = {
             'adhaarModel': YOLO('models/aadhaarYolov8.pt'),
-            'panModel': YOLO('models/PanYolov8.pt')  # Load additional models as needed
+            'panModel': YOLO('models/PanModal_v3.pt')  # Load additional models as needed
         }
 
     return app

app/routes/panApi.py

@@ -55,7 +55,9 @@ def ocrPan(mode, session):
         # Run detection
         model = current_app.models.get('panModel')
         results = model.predict(source=img, save=False)
+        # print(results,"model result")
         extracted_data = process_results(results, img)
+        # print(extracted_data, "extracted data")
         
         if extracted_data.get('statusCode') == 400:
             return jsonify(extracted_data), 400
@@ -67,4 +69,4 @@ def ocrPan(mode, session):
         return jsonify({"error": "Unable to identify image format."}), 400
     except Exception as e:
         current_app.logger.error(f"Unexpected error: {str(e)}")
-        return jsonify({"error": "An unexpected error occurred."}), 500
+        return jsonify({"error": "An unexpected error occurred on api call."}), 500

app/services/panServices/panDataExtractor.py

@@ -1,7 +1,8 @@
 import re
 
 def extract_panData(data):
-    unwanted_words = ["Name","/Name", "Date of Birth", "/Date of Birth", "Permanent Account Number", "Father's Name", "14 /Name", "/Father's Name"]
+    unwanted_words = ["Name","/Name",'Permanent', 'Account', 'Number', 'Card', 'नाम', '/Name',
+        "पिता का नाम",'नाम / Name', "पिता का नाम/ Father's Name", 'नाम /Name',"पिता का नाम / Father's Name", 'जन्म का वाराज़', 'Date of Birth', 'Permanent Account Number Card', "Date of Birth", "/Date of Birth", "Permanent Account Number", "Father's Name", "14 /Name", "/Father's Name"]
     
     # Clean the array by removing unwanted words and invalid entries
     cleaned_data = []
@@ -24,30 +25,30 @@ def extract_panData(data):
     }
     
     # Check and extract PAN number
-    print(cleaned_data)
+    print(cleaned_data, "cleaned data")
     pan_pattern = re.compile(r'^[A-Z]{5}[0-9]{4}[A-Z]$')
-    if len(cleaned_data) > 0 and pan_pattern.match(cleaned_data[3]):
-        result["data"]["panNo"] = cleaned_data[3]
+    if len(cleaned_data) > 0 and pan_pattern.match(cleaned_data[0]):
+        result["data"]["panNo"] = cleaned_data[0]
     else:
         result["data"]["panNo"] = ''
         
     # Check and extract name
     name_pattern = re.compile(r'^[A-Za-z .]+$')
-    if len(cleaned_data) > 1 and name_pattern.match(cleaned_data[2]):
-        result["data"]["name"] = cleaned_data[2]
+    if len(cleaned_data) > 1 and name_pattern.match(cleaned_data[1]):
+        result["data"]["name"] = cleaned_data[1]
     else:
         result["data"]["name"] = ''
         
     # Check and extract father's name
-    if len(cleaned_data) > 2 and name_pattern.match(cleaned_data[1]):
-        result["data"]["fatherName"] = cleaned_data[1]
+    if len(cleaned_data) > 2 and name_pattern.match(cleaned_data[2]):
+        result["data"]["fatherName"] = cleaned_data[2]
     else:
         result["data"]["fatherName"] = ''
         
     # Check and extract date of birth
     dob_pattern = re.compile(r'^\d{2}[-/]\d{2}[-/]\d{4}$')
-    if len(cleaned_data) > 3 and dob_pattern.match(cleaned_data[0]):
-        result["data"]["dob"] = cleaned_data[0]
+    if len(cleaned_data) > 3 and dob_pattern.match(cleaned_data[3]):
+        result["data"]["dob"] = cleaned_data[3]
     else:
         result["data"]["dob"] = ''
     

app/services/panServices/panOcr.py

@@ -2,41 +2,55 @@ from io import BytesIO
 from ...utils.azureOCR import analyze_image
 from ...utils.imageUtils import resize_if_needed, all_cropped_images_to_one_image
 from .panDataExtractor import extract_panData
+from collections import defaultdict
 
 def process_results(results, img):
-    label_indices = {"dob": 0, "father": 1, "name": 2, "pan_num": 3}
+    label_indices = {"pan_num": 0, "name": 1, "father": 2, "dob": 3}
     confidence_threshold = 0.3
     input_image_format = img.format if img.format else "PNG"
     valid_formats = ["JPEG", "PNG", "BMP", "GIF", "TIFF"]
     input_image_format = input_image_format if input_image_format in valid_formats else "PNG"
     
-    cropped_images_with_labels = []
+    best_crops = {label: (None, -1) for label in label_indices.keys()}  # Store best (image, confidence) pairs
+
     precision_data = {label: {"correct": 0, "total": 0} for label in label_indices.keys()}
-    # extracted_data = {"pan_num": "", "name": "", "father": "", "dob": ""}
     
     for result in results:
         for bbox, cls, conf in zip(result.boxes.xyxy, result.boxes.cls, result.boxes.conf):
-            label = ["dob", "father", "name", "pan_num"][int(cls)]
+            # Ensure the class index is within the bounds of the label list
+            if int(cls) >= len(label_indices):
+                print(f"Warning: Class index {cls} is out of range. Skipping this bbox.")
+                continue
+            
+            label = list(label_indices.keys())[int(cls)]
             print(label, conf)
             if conf < confidence_threshold:
                 continue
-            
+
             x1, y1, x2, y2 = map(int, bbox.tolist())
             crop_img = img.crop((x1, y1, x2, y2))
             crop_img = resize_if_needed(crop_img)
-            # crop_img.save(f"temp_{label}.png")
-            cropped_images_with_labels.append((crop_img, label_indices[label], conf))
-            precision_data[label]["total"] += 1
-            precision_data[label]["correct"] += 1  # Replace with actual OCR validation check
+            crop_img.save(f"temp_{label}.png")
+
+            # Replace old crop if new one has higher confidence
+            _, best_conf = best_crops[label]
+            if conf > best_conf:
+                best_crops[label] = (crop_img, conf)
+                precision_data[label]["total"] += 1
+                precision_data[label]["correct"] += 1  # Replace with actual OCR validation check
+
+    # Extract the images for final processing
+    cropped_images_with_labels = [(img, label_indices[label], conf) for label, (img, conf) in best_crops.items() if img is not None]
     
     # Sort the images by their label indices in ascending order
     cropped_images_with_labels.sort(key=lambda x: x[1])
     print(cropped_images_with_labels, "cropped images with labels")
+    
+    if not cropped_images_with_labels:
+        raise ValueError("No images were cropped.")
+
     # Extract only the images for concatenation
     cropped_images = [img for img, _, _ in cropped_images_with_labels]
-    # print(cropped_images, "cropped images")
-    if not cropped_images:
-        raise ValueError("No images were cropped.")
     
     final_image = all_cropped_images_to_one_image(cropped_images, separator_image_path='app/utils/seprator3.png')
     buffer = BytesIO()
@@ -49,4 +63,4 @@ def process_results(results, img):
     texts = [line['text'] for line in lines]
     print(texts, "text after microsoft ocr")
     extracted_data = extract_panData(texts)
-    return extracted_data
+    return extracted_data