Update app.py

app.py

@@ -2,43 +2,57 @@ import gradio as gr
 from paddleocr import PaddleOCR
 from PIL import Image
 import numpy as np
+import os
 
-# Initialize PaddleOCR (English language, with handwriting support enabled)
-# This will download the models on the first run
-ocr = PaddleOCR(use_angle_cls=True, lang='en', rec_algorithm='SVTR_LCNet')
+# Initialize OCR with specific settings for handwriting and forms
+# rec_algorithm='SVTR_LCNet' is the secret for messy handwriting
+ocr = PaddleOCR(
+    use_angle_cls=True, 
+    lang='en', 
+    show_log=False,
+    rec_algorithm='SVTR_LCNet' 
+)
 
 def process_bank_form(image):
     if image is None:
         return "Please upload an image."
 
-    # Convert PIL to numpy array for PaddleOCR
-    img_array = np.array(image.convert("RGB"))
+    # Pre-processing: Convert to RGB and then to a format Paddle understands
+    img = image.convert("RGB")
+    img_array = np.array(img)
     
-    # Run OCR
-    # det=True (finds text), rec=True (reads text), cls=True (handles rotated text)
+    # Run the OCR pipeline
+    # We use 'cls=True' to fix upside-down or tilted phone photos automatically
     result = ocr.ocr(img_array, cls=True)
 
     if not result or result[0] is None:
-        return "No text detected. Try a clearer photo."
+        return "No text found. Try taking a photo with more light or a darker pen."
 
-    # Format the output
-    full_text = []
+    # Extracting and cleaning the text
+    extracted_lines = []
     for line in result[0]:
-        text = line[1][0]      # The actual words
-        confidence = line[1][1] # How sure the AI is
-        if confidence > 0.5:    # Filter out "visual noise"
-            full_text.append(text)
+        text_content = line[1][0]
+        confidence = line[1][1]
+        
+        # Only keep text if the AI is reasonably sure (avoids random symbols)
+        if confidence > 0.4:
+            extracted_lines.append(text_content)
     
-    return "\n".join(full_text)
+    return "\n".join(extracted_lines)
 
-# Build the Interface
-demo = gr.Interface(
-    fn=process_bank_form,
-    inputs=gr.Image(type="pil", label="Upload Bank Form"),
-    outputs=gr.Textbox(label="Extracted Text (Handwritten & Printed)", lines=20),
-    title="🏦 Advanced Bank Form OCR",
-    description="Uses PP-OCRv5 to detect messy handwriting and printed text on a CPU."
-)
+# Create the Gradio Interface
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🏦 Bank Form Handwriting Reader")
+    gr.Markdown("Upload a photo of your bank form. This system is tuned for English handwritten and printed text.")
+    
+    with gr.Row():
+        with gr.Column():
+            input_img = gr.Image(type="pil", label="Bank Form Photo")
+            submit_btn = gr.Button("Read Form", variant="primary")
+        with gr.Column():
+            output_text = gr.Textbox(label="Extracted Text", lines=20, show_copy_button=True)
+            
+    submit_btn.click(fn=process_bank_form, inputs=input_img, outputs=output_text)
 
 if __name__ == "__main__":
     demo.launch()