Extract images from PAGE-XML

extract_images_from_pagexml.py

from pathlib import Path
from lxml import etree
from PIL import Image
import numpy as np
import cv2

PAGE_NS = {
    "page": "http://schema.primaresearch.org/PAGE/gts/pagecontent/2019-07-15"
}


def parse_points(points_string):
    pts = []

    for p in points_string.split():
        x, y = p.split(",")
        pts.append((int(x), int(y)))

    return np.array(pts, dtype=np.int32)


def extract_graphics(xml_file, image_file, output_dir):

    output_dir.mkdir(parents=True, exist_ok=True)

    tree = etree.parse(str(xml_file))

    img = Image.open(image_file).convert("RGB")
    img_np = np.array(img)

    graphic_regions = tree.xpath("//page:GraphicRegion", namespaces=PAGE_NS)

    for i, region in enumerate(graphic_regions):

        region_id = region.get("id", f"region_{i}")

        coords = region.find("page:Coords", namespaces=PAGE_NS)

        if coords is None:
            continue

        polygon = parse_points(coords.get("points"))

        x, y, w, h = cv2.boundingRect(polygon)

        cropped = img_np[y:y+h, x:x+w]

        # shift polygon to cropped coordinates
        shifted = polygon - np.array([[x, y]])

        mask = np.zeros((h, w), dtype=np.uint8)
        cv2.fillPoly(mask, [shifted], 255)

        result = cv2.bitwise_and(cropped, cropped, mask=mask)

        # RGBA output
        rgba = cv2.cvtColor(result, cv2.COLOR_RGB2RGBA)
        rgba[:, :, 3] = mask

        outfile = output_dir / f"{image_file.stem}_{region_id}.png"

        Image.fromarray(rgba).save(outfile)

        print(outfile)


for xml_file in xml_dir.glob("*.xml"):

    stem = xml_file.stem

    for ext in (".tif", ".jpg", ".png"):

        image_file = img_dir / (stem + ext)

        if image_file.exists():
            extract_graphics(xml_file, image_file, out_dir)
            break

    extract_graphics(
        xml_dir = Path("./pagexml"),
        img_dir = Path("./images"),
        out_dir = Path("./graphics")
    )