app.py

import cv2
import os
import pytesseract
import gradio as gr
from gradio import Interface, Image, Text
import numpy as np
from PIL import Image
from PIL import UnidentifiedImageError

def process_image(input_image):
    try:
        # Convert the input image to a NumPy array if it's a PIL Image
        if isinstance(input_image, Image.Image):
            img = np.array(input_image)
        else:
            # If it's a file path or file-like object, read it directly with OpenCV
            img = cv2.imread(input_image)

        # Check that the image is in the expected format
        if img is None or img.dtype != np.uint8:
            raise Exception("Could not read the image. Please check the image format.")

        img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)

        # img = cv2.imdecode(np.fromstring(input_image.read(), np.uint8), cv2.IMREAD_COLOR)
        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        ret, thresh1 = cv2.threshold(gray, 0, 255, cv2.THRESH_OTSU | cv2.THRESH_BINARY_INV)
        rect_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (18, 18))
        dilation = cv2.dilate(thresh1, rect_kernel, iterations=1)

        # Find text lines using connected component analysis
        text_lines = []
        contours, hierarchy = cv2.findContours(dilation, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
        for cnt in contours:
            x, y, w, h = cv2.boundingRect(cnt)
            text_lines.append((y, y + h, x, x + w))

        # Sort text lines by their y-coordinates
        text_lines.sort(key=lambda line: line[0])

        # Extract text from each line using Tesseract
        recognized_text = []
        for y_min, y_max, x_min, x_max in text_lines:
            cropped_img = img[y_min:y_max, x_min:x_max]
            custom_config = r'-l eng+khm --oem 3 --psm 6'
            extracted_text = pytesseract.image_to_string(cropped_img, config=custom_config)
            recognized_text.append(extracted_text.strip())

        # Combine recognized text into a single string
        full_text = "\n".join(recognized_text)

        # Draw bounding boxes on the image
        result_rgb = img.copy()
        for y_min, y_max, x_min, x_max in text_lines:
            cv2.rectangle(result_rgb, (x_min, y_min), (x_max, y_max), (0, 255, 0), 2)

        return full_text, result_rgb
    except Exception as e:
        return "Could not process the image. Error: " + str(e), None

iface = gr.Interface(
    process_image,
    inputs=[gr.Image(type="pil", label="Processed Image")],
    outputs=[
        gr.Text(label="Detected Labels"),
        gr.Image(type="pil", label="Processed Image")
    ],
    title="Bank Statement OCR",
    # description="Upload an image containing text to perform OCR and see the detected text and image."
    flagging_options=["blurry", "incorrect", "other"],)

iface.launch(debug=True , share=True)