app.py

import os
import fitz  # PyMuPDF for PDF handling
from transformers import DonutProcessor, VisionEncoderDecoderModel
from PIL import Image
import tempfile
import streamlit as st


def extract_text_with_donut(pdf_path):
    """
    Extract text using Hugging Face Donut model for OCR.
    :param pdf_path: Path to the input PDF file.
    :return: List of extracted text for each page.
    """
    # Load the model and processor
    processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base")
    model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base")

    extracted_text = []
    doc = fitz.open(pdf_path)

    for page_num in range(len(doc)):
        page = doc.load_page(page_num)
        pix = page.get_pixmap(dpi=300)  # Convert PDF page to high-resolution image
        image_path = f"temp_page_{page_num}.png"
        pix.save(image_path)

        # Perform OCR using Donut
        image = Image.open(image_path).convert("RGB")
        inputs = processor(images=image, return_tensors="pt")
        outputs = model.generate(**inputs)

        page_text = processor.batch_decode(outputs, skip_special_tokens=True)[0]
        extracted_text.append({"page_num": page_num, "text": page_text})

        # Cleanup temporary image
        os.remove(image_path)

    return extracted_text


def overlay_text_with_fonts(pdf_path, extracted_data, output_pdf_path):
    """
    Overlay extracted text onto the original PDF.
    :param pdf_path: Path to the input PDF file.
    :param extracted_data: Extracted text for each page.
    :param output_pdf_path: Path to save the output PDF file.
    """
    doc = fitz.open(pdf_path)

    for item in extracted_data:
        page_num = item["page_num"]
        text = item["text"]

        page = doc[page_num]

        # Add extracted text to the page
        y = 50  # Starting position
        for line in text.split("\n"):
            page.insert_text((50, y), line, fontsize=10, fontname="Helvetica", color=(0, 0, 0))
            y += 12  # Line spacing

    doc.save(output_pdf_path)
    print(f"PDF saved to: {output_pdf_path}")


def process_pdf(uploaded_pdf, output_pdf_path):
    """
    Process the uploaded PDF to extract text using Hugging Face Donut and overlay it.
    :param uploaded_pdf: Uploaded PDF file.
    :param output_pdf_path: Path to save the output PDF file.
    """
    with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as temp_pdf:
        temp_pdf.write(uploaded_pdf.read())
        temp_pdf_path = temp_pdf.name

    extracted_data = extract_text_with_donut(temp_pdf_path)
    overlay_text_with_fonts(temp_pdf_path, extracted_data, output_pdf_path)

    os.remove(temp_pdf_path)


# Streamlit App
def main():
    st.title("Hugging Face OCR Text Extraction Tool")
    st.write("Upload a PDF to extract and overlay text using Hugging Face Donut.")

    uploaded_file = st.file_uploader("Upload PDF", type=["pdf"])
    if uploaded_file:
        output_pdf_path = "converted_output.pdf"

        with st.spinner("Processing your PDF..."):
            process_pdf(uploaded_file, output_pdf_path)

        st.success("PDF processing complete!")

        # Provide a download button for the processed PDF
        with open(output_pdf_path, "rb") as f:
            st.download_button(
                label="Download Converted PDF",
                data=f,
                file_name="converted_output.pdf",
                mime="application/pdf"
            )

        os.remove(output_pdf_path)


if __name__ == "__main__":
    main()