import gradio as gr import fitz # PyMuPDF import cv2 from pdf2image import convert_from_path import pytesseract from pytesseract import Output import numpy as np import os from fpdf import FPDF import difflib # For text comparison # Convert PDFs to images def convert_pdf_to_images(pdf_path, dpi=300): images = convert_from_path(pdf_path, dpi=dpi, poppler_path="/usr/bin") return [cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR) for image in images] # Align images def align_images(img1, img2): gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY) gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY) orb = cv2.ORB_create() kp1, des1 = orb.detectAndCompute(gray1, None) kp2, des2 = orb.detectAndCompute(gray2, None) bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True) matches = bf.match(des1, des2) matches = sorted(matches, key=lambda x: x.distance) src_pts = np.float32([kp1[m.queryIdx].pt for m in matches]).reshape(-1, 1, 2) dst_pts = np.float32([kp2[m.trainIdx].pt for m in matches]).reshape(-1, 1, 2) matrix, _ = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0) # Validate if alignment is good enough if matrix is None or len(matches) < 10: # Check if sufficient matches exist raise ValueError("Alignment failed. Insufficient matches between images.") aligned_img = cv2.warpPerspective(img2, matrix, (img1.shape[1], img1.shape[0])) return aligned_img # Compare visual changes def compare_visual_changes(orig_img, edit_img, start_position): diff = cv2.absdiff(orig_img, edit_img) gray_diff = cv2.cvtColor(diff, cv2.COLOR_BGR2GRAY) # Apply Gaussian blur to reduce noise blurred_diff = cv2.GaussianBlur(gray_diff, (5, 5), 0) # Apply thresholding _, thresh = cv2.threshold(blurred_diff, 70, 255, cv2.THRESH_BINARY) # Morphological operations to clean noise kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5)) cleaned = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel) contours, _ = cv2.findContours(cleaned, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) overlay = edit_img.copy() visual_changes = [] position_counter = start_position font = cv2.FONT_HERSHEY_SIMPLEX font_scale = 0.8 thickness = 2 for cnt in contours: if cv2.contourArea(cnt) > 100: # Filter out small regions x, y, w, h = cv2.boundingRect(cnt) cv2.rectangle(overlay, (x, y), (x + w, y + h), (0, 0, 255), 2) # Red bounding box cv2.putText(overlay, str(position_counter), (x, y - 10), font, font_scale, (0, 255, 0), thickness) visual_changes.append((position_counter, f'Visual change detected at position {position_counter}')) position_counter += 1 return overlay, visual_changes, position_counter # Normalize and clean text to reduce noise def normalize_text(text): return text.strip().lower() # Convert to lower case and remove leading/trailing spaces # Compare text changes with bounding boxes with normalization def compare_text_changes_with_boxes(orig_img, edit_img, start_position): # Set Tesseract configuration options custom_config = r'--oem 3 --psm 4' orig_data = pytesseract.image_to_data(orig_img, output_type=Output.DICT, config=custom_config) edit_data = pytesseract.image_to_data(edit_img, output_type=Output.DICT, config=custom_config) orig_text = [normalize_text(t) for t in orig_data['text']] edit_text = [normalize_text(t) for t in edit_data['text']] diff = difflib.ndiff(orig_text, edit_text) overlay = edit_img.copy() text_changes = [] position_counter = start_position font = cv2.FONT_HERSHEY_SIMPLEX font_scale = 0.8 thickness = 2 for line in diff: if line.startswith("+ "): # Added text text = line[2:].strip() if text and text in edit_data['text']: index = edit_data['text'].index(text) x, y, w, h = edit_data['left'][index], edit_data['top'][index], edit_data['width'][index], edit_data['height'][index] cv2.rectangle(overlay, (x, y), (x + w, y + h), (0, 0, 255), 2) cv2.putText(overlay, str(position_counter), (x, y - 10), font, font_scale, (0, 255, 0), thickness) text_changes.append((position_counter, f'"{text}" added at position {position_counter}')) position_counter += 1 elif line.startswith("- "): # Removed text text = line[2:].strip() if text and text in orig_data['text']: index = orig_data['text'].index(text) x, y, w, h = orig_data['left'][index], orig_data['top'][index], orig_data['width'][index], orig_data['height'][index] cv2.rectangle(overlay, (x, y), (x + w, y + h), (0, 0, 255), 2) cv2.putText(overlay, str(position_counter), (x, y - 10), font, font_scale, (0, 255, 0), thickness) text_changes.append((position_counter, f'"{text}" removed at position {position_counter}')) position_counter += 1 return overlay, text_changes, position_counter # Sanitize text for PDF compatibility def sanitize_text(text): return text.encode('latin-1', errors='replace').decode('latin-1') # Generate PDF report def generate_report(images, changes, title, output_path): pdf = FPDF() for img in images: temp_path = "temp_image.png" cv2.imwrite(temp_path, img) pdf.add_page() pdf.image(temp_path, x=10, y=10, w=190) os.remove(temp_path) pdf.add_page() pdf.set_font("Arial", size=12) pdf.cell(0, 10, sanitize_text(title), ln=True, align="C") pdf.ln(10) for _, change in changes: pdf.cell(0, 10, sanitize_text(change), ln=True) pdf.output(output_path) return output_path # Perform visual and text comparisons separately def generate_separate_comparisons(original_pdf, edited_pdf): original_images = convert_pdf_to_images(original_pdf) edited_images = convert_pdf_to_images(edited_pdf) # Visual comparison visual_combined_images = [] visual_changes = [] position_counter = 1 for orig_img, edit_img in zip(original_images, edited_images): aligned_img = align_images(orig_img, edit_img) highlighted_img, page_visual_changes, position_counter = compare_visual_changes( orig_img, aligned_img, position_counter ) visual_changes.extend(page_visual_changes) visual_combined_images.append(np.hstack((orig_img, highlighted_img))) # Generate visual changes report visual_report_path = generate_report( visual_combined_images, visual_changes, "Visual Changes", "outputs/visual_changes.pdf" ) # Text comparison text_combined_images = [] text_changes = [] position_counter = 1 for orig_img, edit_img in zip(original_images, edited_images): aligned_img = align_images(orig_img, edit_img) highlighted_img, page_text_changes, position_counter = compare_text_changes_with_boxes( orig_img, aligned_img, position_counter ) text_changes.extend(page_text_changes) text_combined_images.append(np.hstack((orig_img, highlighted_img))) # Generate text changes report text_report_path = generate_report( text_combined_images, text_changes, "Text Changes", "outputs/text_changes.pdf" ) return visual_report_path, text_report_path # Gradio interface function def pdf_comparison(original_pdf, edited_pdf): visual_path, text_path = generate_separate_comparisons(original_pdf.name, edited_pdf.name) return visual_path, text_path # Gradio interface interface = gr.Interface( fn=pdf_comparison, inputs=[ gr.File(label="Upload Original PDF", file_types=[".pdf"]), gr.File(label="Upload Edited PDF", file_types=[".pdf"]) ], outputs=[ gr.File(label="Download Visual Changes Report"), gr.File(label="Download Text Changes Report") ], title="PDF Comparison Tool with Separate Comparisons", description="Upload two PDFs: the original and the edited version. The tool generates separate reports for visual and text changes." ) if __name__ == "__main__": interface.launch()