invert black parts in page

gistfile1.txt

import fitz  # PyMuPDF
import cv2
import numpy as np
from PIL import Image, ImageOps
import os

# -----------------------------
# INPUT/OUTPUT
# -----------------------------
input_pdf = "input.pdf"
output_pdf = "output_fixed.pdf"

# Temporary folder to store images
tmp_folder = "tmp_pages"
os.makedirs(tmp_folder, exist_ok=True)

# -----------------------------
# PARAMETERS
# -----------------------------
zoom = 3               # 3x zoom (~216 dpi, good clarity without huge size)
threshold_value = 40   # Detect dark pixels
min_area = 20000       # Ignore small regions (text, thin lines)
dark_ratio_min = 0.6   # Only invert if >60% dark pixels
jpeg_quality = 90      # Save images as JPEG to reduce size

# -----------------------------
# OPEN PDF
# -----------------------------
doc = fitz.open(input_pdf)
new_doc = fitz.open()

for page_num in range(len(doc)):
    page = doc[page_num]

    # Render page at high DPI
    mat = fitz.Matrix(zoom, zoom)
    pix = page.get_pixmap(matrix=mat, alpha=False)

    # Convert to numpy image
    img = np.frombuffer(pix.samples, dtype=np.uint8).reshape(pix.height, pix.width, pix.n)
    gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

    # Threshold: detect dark regions
    _, mask = cv2.threshold(gray, threshold_value, 255, cv2.THRESH_BINARY_INV)

    # Find contours of dark regions
    contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

    pil_img = Image.fromarray(img)

    for cnt in contours:
        x, y, w, h = cv2.boundingRect(cnt)

        # Ignore small regions (text, thin lines, borders)
        if w * h < min_area:
            continue

        # Extract region as numpy
        region_np = np.array(pil_img.crop((x, y, x + w, y + h)))
        gray_region = cv2.cvtColor(region_np, cv2.COLOR_RGB2GRAY)

        # Calculate darkness ratio
        dark_pixels = np.sum(gray_region < threshold_value)
        total_pixels = gray_region.size
        dark_ratio = dark_pixels / total_pixels

        # Only invert if mostly dark (solid black block)
        if dark_ratio > dark_ratio_min:
            region = pil_img.crop((x, y, x + w, y + h))
            inverted = ImageOps.invert(region)
            pil_img.paste(inverted, (x, y))

    # Save page as JPEG to reduce file size
    temp_path = os.path.join(tmp_folder, f"page_{page_num+1}.jpg")
    pil_img.save(temp_path, quality=jpeg_quality, dpi=(zoom*72, zoom*72))

    # Insert JPEG back into PDF
    new_page = new_doc.new_page(width=page.rect.width, height=page.rect.height)
    new_page.insert_image(page.rect, filename=temp_path)

# Save final compressed PDF
new_doc.save(output_pdf)
print("✅ Done! Fixed PDF saved as:", output_pdf)