I’m trying to extract data from a PDF form that includes text fields, checkboxes and other input elements. The form looks like a structured intake form.

1. PyMuPDF

    def extract_text_from_pdf(pdf_path):
    with fitz.open(pdf_path) as pdf_file:
        all_text = ""
        for page_num in range(pdf_file.page_count):
            page = pdf_file[page_num]
            all_text += page.get_text()
    return all_text
   

Extracts raw text.But No information about form fields or checkboxes and also loses structure of the form.

2. Opencv

    def convert_pdf_to_images(pdf_path):
        return convert_from_path(pdf_path)
   
    # Preprocess image for better conthattour detection
    def preprocess_image(image):
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        _, binary = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY_INV)
        kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
        dilated = cv2.dilate(binary, kernel, iterations=1)
        return dilated
   
    # Detect main and sub boxes and extract text immediately
    def detect_and_extract_text(image, original_image):
        contours, _ = cv2.findContours(image, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
        extracted_data = []
   
        # print("\nExtracting text from bounding boxes:\n" + "-"*50)
        for contour in contours:
            x, y, w, h = cv2.boundingRect(contour)
            # Filter small boxes
            if w > 50 and 50 < h < 1000:
                # Draw the box on the original image
                cv2.rectangle(original_image, (x, y), (x + w, y + h), (0, 255, 0), 2)
   
                # Extract the region of interest (ROI)
                roi = original_image[y:y + h, x:x + w]
   
                # Extract text using Tesseract
                text = pytesseract.image_to_string(roi, config="--psm 6").strip()
   
                # Append data to the list
                extracted_data.append({"text": text, "bbox": (x, y, w, h)})
   
                # Print the text for this bounding box
                # print(f"Text: {text}")
                # print(f"Bounding Box: {x, y, w, h}")
                # print("-" * 50)
   
        return extracted_data, original_image
   
    # Global OCR for remaining text excluding bounding box areas
    def extract_remaining_text_excluding_boxes(image, bounding_boxes):
        # Create a mask to exclude bounding box areas
        mask = np.ones(image.shape[:2], dtype="uint8") * 255  # Start with a white mask
        for box in bounding_boxes:
            x, y, w, h = box
            cv2.rectangle(mask, (x, y), (x + w, y + h), 0, -1)  # Black out bounding box area
   
        # Apply the mask to the image
        masked_image = cv2.bitwise_and(image, image, mask=mask)
   
        # Convert the masked image to grayscale
        gray_image = cv2.cvtColor(masked_image, cv2.COLOR_BGR2GRAY)
   
        # Perform OCR on the masked image
        text = pytesseract.image_to_string(gray_image, config="--psm 6").strip()
   
        # print("\nExtracting text from the remaining areas (global OCR):\n" + "-"*50)
        # print(f"Remaining Text:\n{text}")
        # print("-" * 50)
   
        return text
   
    # Main execution function
    def process_pdf_and_detect_boxes(pdf_path, output_image_path):
        # Step 1: Convert PDF to images
        pages = convert_pdf_to_images(pdf_path)
        all_extracted_data = []
   
        for i, page in enumerate(pages):
            # print(f"\nProcessing page {i + 1}...\n" + "="*50)
   
            # Convert PIL Image to OpenCV format
            original_image = cv2.cvtColor(np.array(page), cv2.COLOR_RGB2BGR)
   
            # Step 2: Preprocess the image
            preprocessed_image = preprocess_image(original_image)
   
            # Step 3: Detect and extract text box by box
            extracted_data, image_with_boxes = detect_and_extract_text(preprocessed_image, original_image)
            all_extracted_data.extend(extracted_data)
   
            # Collect bounding boxes for exclusion in global OCR
            bounding_boxes = [entry["bbox"] for entry in extracted_data]
   
            # Step 4: Perform global OCR to extract remaining text
            remaining_text = extract_remaining_text_excluding_boxes(original_image, bounding_boxes)
            if remaining_text:
                all_extracted_data.append({"text": remaining_text, "bbox": None})
   
            # Step 5: Save image with bounding boxes
            cv2.imwrite(f"{output_image_path}_page_{i + 1}.png", image_with_boxes)
            # print(f"Saved image with boxes: {output_image_path}_page_{i + 1}.png")
   

Works well for extracting text from images. But unable to accurately detect checkboxes. And fails to maintain field-value pairs.

3. pdf2docx Lost form structure and checkboxes and resulting data was difficult to parse.

So, how can I extract form field values and checkbox states from a PDF using Python? Are there any libraries or techniques that handle checkboxes and structured form data effectively?

Also attached the PDF for your reference. PDF: PDF_flile