pdf.py

from pdf2image import convert_from_path
from paddleocr import PaddleOCR
from reportlab.pdfgen import canvas
from PyPDF2 import PdfReader, PdfWriter
import io
import numpy as np
import cv2
from tqdm import tqdm

def process_page(page_num, pil_image, input_pdf_path):
    if pil_image.mode != 'RGB':
        pil_image = pil_image.convert('RGB')

    # Convert PIL Image to a format compatible with PaddleOCR (which uses OpenCV)
    image = np.array(pil_image)
    # OpenCV expects images in BGR format
    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

    # Perform OCR
    ocr_results = model.ocr(image, cls=True)

    # Create a PDF page with text overlay
    packet = io.BytesIO()
    can = canvas.Canvas(packet)

    for line in ocr_results:
        for element in line:
            bbox, (text, _) = element
            x, y = bbox[0]
            # Use image.shape[0] for the image height
            can.drawString(x, image.shape[0] - y, text)

    can.save()

    # Move the "canvas" into a PDF object
    packet.seek(0)
    new_pdf = PdfReader(packet)
    page = new_pdf.pages[0]

    # Merge with original PDF page
    original_pdf = PdfReader(input_pdf_path)
    page.merge_page(original_pdf.pages[page_num])

    return page

def ocr_pdf_to_text_pdf(input_pdf_path, output_pdf_path):
    # Convert PDF to images
    images = convert_from_path(input_pdf_path)

    # Create a PDF writer object
    output_pdf = PdfWriter()

    for i, image in tqdm(enumerate(images), total=len(images)):
        page = process_page(i, image, input_pdf_path)
        output_pdf.add_page(page)

    # Write the output PDF to a file
    with open(output_pdf_path, "wb") as output_file:
        output_pdf.write(output_file)

# Usage
input_pdf_path = 'SOC VOL. 3.pdf'
output_pdf_path = 'OCR_SOC VOL. 3.pdf'
ocr_pdf_to_text_pdf(input_pdf_path, output_pdf_path)