app.py

import os
import uuid
from flask import Flask, request, jsonify
from flask_cors import CORS
from werkzeug.utils import secure_filename
import torch
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
import cv2
import numpy as np

app = Flask(__name__)
CORS(app)

# Configure upload folder
UPLOAD_FOLDER = 'uploads'
ALLOWED_EXTENSIONS = {'png', 'jpg', 'jpeg', 'pdf', 'tif', 'tiff'}
app.config['UPLOAD_FOLDER'] = UPLOAD_FOLDER
app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024  # 16MB max upload

# Create uploads directory if it doesn't exist
os.makedirs(UPLOAD_FOLDER, exist_ok=True)

# Load OCR model - Microsoft's TrOCR model
print("Loading OCR model...")
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-handwritten", cache_dir="/huggingface_cache")
model = VisionEncoderDecoderModel.from_pretrained("microsoft/trocr-base-handwritten", cache_dir="/huggingface_cache")

# Move model to GPU if available
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")
model.to(device)

def allowed_file(filename):
    return '.' in filename and \
           filename.rsplit('.', 1)[1].lower() in ALLOWED_EXTENSIONS

def preprocess_image(image_path):
    # Open image with PIL
    image = Image.open(image_path).convert("RGB")
    
    # Basic enhancement for better OCR results
    # Convert to OpenCV format for preprocessing
    img = np.array(image)
    img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
    
    # Apply adaptive thresholding to handle varying lighting conditions
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, 
                                  cv2.THRESH_BINARY, 11, 2)
    
    # Convert back to PIL
    enhanced_image = Image.fromarray(cv2.cvtColor(thresh, cv2.COLOR_GRAY2RGB))
    return enhanced_image

def perform_ocr(image_path):
    # Preprocess the image
    image = preprocess_image(image_path)
    
    # Prepare image for the model
    pixel_values = processor(image, return_tensors="pt").pixel_values.to(device)
    
    # Generate text
    generated_ids = model.generate(
        pixel_values,
        max_length=64,
        num_beams=5,
        early_stopping=True
    )
    
    # Decode generated text
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
    return generated_text.strip()

@app.route('/ocr', methods=['POST'])
def ocr():
    # Check if a file was uploaded
    if 'file' not in request.files:
        return jsonify({'error': 'No file part'}), 400
    
    file = request.files['file']
    
    # Check if filename is empty
    if file.filename == '':
        return jsonify({'error': 'No selected file'}), 400
    
    # Check if file type is allowed
    if file and allowed_file(file.filename):
        # Create a unique filename
        filename = str(uuid.uuid4()) + '_' + secure_filename(file.filename)
        file_path = os.path.join(app.config['UPLOAD_FOLDER'], filename)
        
        # Save the file
        file.save(file_path)
        
        try:
            # Perform OCR
            extracted_text = perform_ocr(file_path)
            
            # Clean up the file after processing
            os.remove(file_path)
            
            return jsonify({
                'success': True,
                'text': extracted_text
            })
        except Exception as e:
            # Log the error
            print(f"Error processing image: {str(e)}")
            return jsonify({
                'success': False,
                'error': str(e)
            }), 500
    else:
        return jsonify({'error': 'File type not allowed'}), 400

@app.route('/health', methods=['GET'])
def health_check():
    return jsonify({'status': 'healthy'}), 200

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=5000, debug=False)