PDF rendering with PDFBox, from Python

pdfbox.py

# SPDX-FileCopyrightText: 2023 mara004
# SPDX-License-Identifier: CC-BY-4.0 OR Apache-2.0

# Assuming you have an Apache PDFBox 3 jar in the same directory

from pathlib import Path
import jpype
import jpype.imports
import PIL.Image

RELATIVE_DIR = Path(__file__).resolve().parent
PDFBOX_JAR = RELATIVE_DIR / "pdfbox.jar"
jpype.addClassPath(PDFBOX_JAR)
jpype.startJVM("-Djava.awt.headless=true")

import java.io as jio
import org.apache.pdfbox as pdfbox
from org.apache.pdfbox.rendering import ImageType


DPI = 300
TEST_FILE = Path("~/projects/scripts/out/38.pdf").expanduser()
OUTPUT_DIR = RELATIVE_DIR / "out"
OUTPUT_DIR.mkdir(exist_ok=True)


ImageTypeToPIL = {
    ImageType.BINARY: "1",     # TYPE_BYTE_BINARY
    ImageType.GRAY:   "L",     # TYPE_BYTE_GRAY
    ImageType.RGB:    "BGRX",  # TYPE_INT_RGB, actually BGRX in memory
    ImageType.ARGB:   "BGRA",  # TYPE_INT_ARGB, actually BGRA in memory
    ImageType.BGR:    "BGR",   # TYPE_3BYTE_BGR
}


pdf = pdfbox.Loader.loadPDF( jio.File(str(TEST_FILE)) )
renderer = pdfbox.rendering.PDFRenderer(pdf)
n_pages = int( pdf.getNumberOfPages() )

for i in range(n_pages):
    
    print(f"Rendering page {i+1} ...")
    
    imgtype = ImageType.BGR
    pil_srcmode = ImageTypeToPIL[imgtype]
    pil_dstmode = pil_srcmode.replace("BGR", "RGB")
    
    j_image = renderer.renderImageWithDPI(i, DPI, imgtype)
    w, h = int(j_image.getWidth()), int(j_image.getHeight())
    
    j_data = j_image.getRaster().getDataBuffer().getData()
    py_data = memoryview(j_data)  # requires PIL >= 9.5
    
    py_image = PIL.Image.frombuffer(pil_dstmode, (w, h), py_data, "raw", pil_srcmode, 0, 1)
    py_image.save(OUTPUT_DIR / f"render_{i+1}.jpg")


jpype.shutdownJVM()

Actually, I see now there has been some independent discussion on this already, and a kind of similar gist written by @lebedov:
lebedov/python-pdfbox#10
https://gist.github.com/lebedov/cefbec588c2b2bd0251ba505bd8bf933
https://gist.github.com/lebedov/3518142a5c2431b8c9a28d323100558a

Our code above uses a slightly different approach for data transfer, though (accessing the underlying buffer directly instead of calling getRGB()), and the color space handling is new, too. Moreover, we use the newer v3 API of pdfbox, while the previous examples were yet written with pdfbox v2.

todo: might theoretically want to handle offset, though it always seems to be 0 in practice, as per my testing

also want to look into different input types (bytes, byte buffer), and parallelization, as would be relevant in a generic API