ValueError: cannot add non-opaque RGBA color to RGB palette

[Jenztsch]

Hi everyone,

I am using Pillow from Paperless NGX via transitive dependencies. When OCRing some specific PDF files, I get the following stack trace:

Traceback (most recent call last):
  File "/usr/src/paperless/src/paperless_tesseract/parsers.py", line 368, in parse
    ocrmypdf.ocr(**args)
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/api.py", line 375, in ocr
    return run_pipeline(options=options, plugin_manager=plugin_manager)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/_pipelines/ocr.py", line 225, in run_pipeline
    return _run_pipeline(options, plugin_manager)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/_pipelines/ocr.py", line 176, in _run_pipeline
    pdfinfo = get_pdfinfo(
              ^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/_pipeline.py", line 173, in get_pdfinfo
    return PdfInfo(
           ^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 1103, in __init__
    self._pages = _pdf_pageinfo_concurrent(
                  ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 778, in _pdf_pageinfo_concurrent
    executor(
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/_concurrent.py", line 74, in __call__
    self._execute(
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/builtin_plugins/concurrency.py", line 141, in _execute
    result = future.result()
             ^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 449, in result
    return self.__get_result()
           ^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/concurrent/futures/_base.py", line 401, in __get_result
    raise self._exception
  File "/usr/local/lib/python3.11/concurrent/futures/thread.py", line 58, in run
    result = self.fn(*self.args, **self.kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 727, in _pdf_pageinfo_sync
    return PageInfo(pdf, pageno, infile, check_pages, detailed_analysis)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 842, in __init__
    self._gather_pageinfo(pdf, pageno, infile, check_pages, detailed_analysis)
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 887, in _gather_pageinfo
    for info in _process_content_streams(
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 638, in _process_content_streams
    yield from _find_regular_images(container, contentsinfo)
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 554, in _find_regular_images
    yield ImageInfo(name=draw.name, pdfimage=pdfimage, shorthand=draw.shorthand)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/ocrmypdf/pdfinfo/info.py", line 364, in __init__
    pim = PdfImage(pdfimage)
          ^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/pikepdf/models/image.py", line 831, in __init__
    self._jpxpil = self.as_pil_image()
                   ^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/pikepdf/models/image.py", line 740, in as_pil_image
    return Image.open(bio)
           ^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/PIL/Image.py", line 3323, in open
    im = _open_core(
         ^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/PIL/Image.py", line 3304, in _open_core
    im = factory(fp, filename)
         ^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/PIL/ImageFile.py", line 137, in __init__
    self._open()
  File "/usr/local/lib/python3.11/site-packages/PIL/Jpeg2KImagePlugin.py", line 224, in _open
    header = _parse_jp2_header(self.fp)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/PIL/Jpeg2KImagePlugin.py", line 185, in _parse_jp2_header
    palette.getcolor(header.read_fields(">" + ("B" * npc)))
  File "/usr/local/lib/python3.11/site-packages/PIL/ImagePalette.py", line 144, in getcolor
    raise ValueError(msg)
ValueError: cannot add non-opaque RGBA color to RGB palette

I tested the same PDF file with multiple versions of Pillow via calls from OCRmyPDF and Paperless NGX itself. With Pillow 9.4.0 I can successfully OCR the file, while 10.3.0 (the origin of the above stack trace) and even 10.4.0 are raising this error.

Unfortunately I am not sure if the issue is even caused from here. The maintainers of Paperless NGX at least see this as a downstream issue (see here).

Is anyone else familiar with this issue or has any idea where to look for it? Thanks in advance.

[radarhere]

Are you able to modify your copy of pikepdf so that it prints bio.getvalue(), and provide us with the value of that?

[Jenztsch]

Unfortunately I'm not that familiar with Python and my setup is running in a prebuilt Docker container, so I'm not really sure on how to do that from the tip of my tongue. I'll try to check whether I can achieve this in a timely manner and post the data.

At least I can attach one of several example files where this error occurs.

EN_M15_QckStrtBklt_LR_Crop.pdf

[Jenztsch]

I managed to add a simple file writing before this line in the latest pikepdf commit:

with (open(<filename>, 'a+') as writer:
    writer.write(f'bio.getvalue(): {bio.getvalue()}\n')

After one try of adding the above PDF to Paperless, the resulting file is this:

pikepdf.log

[radarhere]

Thanks. I've created #8256 to fix this.

Without it, opening those two images raises the error you mentioned.
With it, the errors do not appear.

[Jenztsch]

That's great news, thank you very much for the effort and the quick response!

[Jenztsch]

Hi, I hope it isn't inconsiderate to ask, but is there an update for this topic? I found more PDFs that produce this error and other users are reporting the same issue. If there is a need, is there something to support/contribute for this?

[radarhere]

Hi. No problem. According to our release schedule, the next Pillow release will be on October 15. Based on past experience, I expect that #8256 will be merged and be part of that release.

If this were just a problem for you, I could build new Pillow wheels for you to install, but you're talking about other users as well.

Instead, here's a way that you could patch Pillow in your code for the moment - try inserting this before you start the OCR process. It should fix your issue.

from PIL import ImagePalette

original_getcolor = ImagePalette.ImagePalette.getcolor

def getcolor(self, color, image=None):
	if len(color) == 4 and not self.colors:
		self.mode = "RGBA"
	return original_getcolor(self, color, image)

ImagePalette.ImagePalette.getcolor = getcolor

Then after October 15, you should be able to remove this workaround and just ask people to upgrade to the latest version of Pillow.

[radarhere]

#8256 has now been merged, so it is just a matter of waiting for the next release on October 15.