app.py

# -*- coding: utf-8 -*-
"""
Final WebApp using Gradio
"""

# Required Libraries
import cv2
import torch
from pytesseract import pytesseract
from transformers import AutoModel, AutoTokenizer
import gradio as gr
import tempfile
import os

# Check if GPU is available
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Load models for OCR
tokenizer_eng = AutoTokenizer.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True)
model_eng = AutoModel.from_pretrained('ucaslcl/GOT-OCR2_0', trust_remote_code=True).to(device).eval()

# Tesseract configuration for Hindi OCR
pytesseract.tesseract_cmd = '/usr/bin/tesseract'
tesseract_config = '--oem 3 --psm 6 -l hin'

# OCR function for both English and Hindi
def perform_ocr(img, language):
    # Use a temporary file for the uploaded image
    with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp_img:
        img.save(temp_img.name)
        img_path = temp_img.name

    res_eng = ""
    res_hin = ""

    if language in ["English", "Both"]:
        # Ensure that inference is done on the correct device (GPU or CPU)
        with torch.no_grad():
            res_eng = model_eng.chat(tokenizer_eng, img_path, ocr_type='ocr')

    if language in ["Hindi", "Both"]:
        img_cv = cv2.imread(img_path)
        res_hin = pytesseract.image_to_string(img_cv, config=tesseract_config)

    # Cleanup temporary file
    os.remove(img_path)

    return res_eng, res_hin

# Keyword Search Functionality
def ocr_and_search(image, language, keyword):
    english_text, hindi_text = perform_ocr(image, language)

    extracted_english = f"Extracted English Text:\n{english_text}" if english_text else "No English text extracted."
    extracted_hindi = f"Extracted Hindi Text:\n{hindi_text}" if hindi_text else "No Hindi text extracted."

    # Search for the keyword in the extracted text
    search_results = []
    if keyword:
        if language in ["English", "Both"] and keyword.lower() in english_text.lower():
            search_results.append(f"Keyword '{keyword}' found in English text.")
        if language in ["Hindi", "Both"] and keyword.lower() in hindi_text.lower():
            search_results.append(f"Keyword '{keyword}' found in Hindi text.")

    search_output = "\n".join(search_results) if search_results else "No matches found."

    return extracted_english, extracted_hindi, search_output

# Gradio Interface Setup
with gr.Blocks() as app:
    gr.Markdown("### OCR Application")
    image_input = gr.Image(type="pil", label="Upload Image")
    language_selection = gr.Radio(choices=["English", "Hindi", "Both"], label="Select Language")
    keyword_input = gr.Textbox(placeholder="Enter keyword to search", label="Keyword Search")
    output_english = gr.Textbox(label="Extracted English Text", interactive=False)
    output_hindi = gr.Textbox(label="Extracted Hindi Text", interactive=False)
    output_search = gr.Textbox(label="Search Results", interactive=False)

    submit_button = gr.Button("Submit")
    submit_button.click(fn=ocr_and_search, inputs=[image_input, language_selection, keyword_input], outputs=[output_english, output_hindi, output_search])

# Launch the Gradio app
app.launch()