pdf2image outputs 1x1 blank image

[dougsouza]

Describe the bug
For some pdf files, convert_from_path, convert_from_bytes outputs a blank 1x1 PIL image. Interestingly for very similar pdfs it works fine. The documents are mostly one very long page pdfs. Any ideais?

To Reproduce
Steps to reproduce the behavior:

1. Unfortunately the pdfs I'm working on are confidential and I am not allowed to share

Expected behavior
I would expect to see a normal PIL image, as happened to other similar pdfs.

Screenshots
Output: [<PIL.PpmImagePlugin.PpmImageFile image mode=RGB size=1x1 at 0x7FCC6C3FA4A8>]. The number of pages is correct, the pdf is just one very long page.

Desktop (please complete the following information):

• OS: Ubuntu 16.04

[Belval]

As pdf2image is only a thin wrapper around pdftoppm, I would try directly from the CLI.

Something like pdftoppm -r 200 -jpeg your_pdf.pdf out.jpg

If you still get a 1x1 pixel then the problem is on their side and I can't help much.

It would also help if you could provide the exact call you do on convert_from_path/convert_from_bytes.

[dougsouza]

@Belval, I figured this is on the pdftoppm side, I am already investigating.

When a run pdftoppm I get: Bogus memory allocation size

The output is a 1x1 blank image.

Thanks for the help

[dougsouza]

@Belval,

Just to share the solution: the mediaBox of my pdf was huge and it was eating up all the memory available rendering a lot of blank space. Turns out that the cropBox of the pdf was correct, so converting using the cropBox (pdftoppm -cropbox ...) works fine.

It would be nice to have an option to use the crop box instead, like:

convert_from_path(..., use_crop_box=False)

Cheers

[Belval]

Can you try with support-cropbox and see if it fixes your issue?

I will upload the new package to PyPi tonight if it does.

[dougsouza]

I just tested. It works fine.

Thanks!!

[Belval]

Pull request merged: #35

Package uploaded: https://pypi.org/project/pdf2image/

Be aware that PyPi caches packages so it can take a few minutes until it is available.

[Belval]

As this was fixed in a previous version, I am closing it.

[alvercau]

Hi

I've run into the same issue. Running with pdftoppm, without the wrapper, does not have the same issue, the jpeg gets correctly generated.
Example file is attached.
5ccad715074ce2b850f646d5-e70d52df-5671-4c97-bfcf-f7e20f5a61d6-1-MN5PADLAME2TMP2HM3JPETQQ33P7NDSQ.pdf

When opening the image afterwards with PIL, I get a DecompressionBombError. So it might be the case that the image is simply too big to be processed by pdf2image.

[Belval]

I was unable to reproduce the issue described with the linked file. Using pdftoppm version 0.62.0 the output image seems to correspond to the PDF.

Here is my code snippet:

from pdf2image import convert_from_path
convert_from_path("test.pdf", size=(3000,))[0].save("out.png")

I changed the name of your PDF for reading purposes, I set a size to avoid the Pillow decompression bomb check.

Could you provide your pdftoppm version with pdftoppm -vand your function call?

[alvercau]

Hi, I didn't set any size, so the issue is indeed that the image is too big. Would it be possible to have convert_from_path raise an error, explaining that the size is too big and that it can be avoided by setting the size parameter, instead of simply silently returning a white pixel?

[Belval]

That's the thing, the underlying library that parses the image file should and does raise an exception on the PDF you linked:

>>> from pdf2image import convert_from_path
>>> convert_from_path("test.pdf")[0].save("out.png")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pdf2image/pdf2image.py", line 202, in convert_from_path
    images += parse_buffer_func(data)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/pdf2image/parsers.py", line 21, in parse_buffer_to_ppm
    images.append(Image.open(BytesIO(data[index : index + file_size])))
  File "/home/ubuntu/.local/lib/python3.6/site-packages/PIL/Image.py", line 2881, in open
    im = _open_core(fp, filename, prefix)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/PIL/Image.py", line 2868, in _open_core
    _decompression_bomb_check(im.size)
  File "/home/ubuntu/.local/lib/python3.6/site-packages/PIL/Image.py", line 2793, in _decompression_bomb_check
    "could be decompression bomb DOS attack." % (pixels, 2 * MAX_IMAGE_PIXELS)
PIL.Image.DecompressionBombError: Image size (382369975 pixels) exceeds limit of 178956970 pixels, could be decompression bomb DOS attack.

Do you override de Pillow limit somewhere else in your code?

[alvercau]

The issue is not that pdf2image returns an image that is too big, but that it returns one single white pixel, without any warnings.
assert convert_from_path("test.pdf")[0].shape == (1,1)
does not give any errors, as the shape of the image is 1.
If I convert the pdf to jpeg by means of the command line tool pdftoppm, the image gets correctly generated, and I do get the DecompressionBomb error when trying to open it with PIL. So the issue is in pdf2image, not in pdftoppm or PIL.

[Belval]

I understand that, but I am simply to reproduce the issue, if I disable Pillow's warning, the saved image is correctly rendered.

In other word, I am unable to get the 1x1 white pixel output you describe.

[mananshah1403]

@Belval I am still encountering this issue with the following PDF. please help!
fail3.pdf

[asanaa8]

same issue here, Help.

[hedes1992]

@Belval,

Just to share the solution: the mediaBox of my pdf was huge and it was eating up all the memory available rendering a lot of blank space. Turns out that the cropBox of the pdf was correct, so converting using the cropBox (pdftoppm -cropbox ...) works fine.

It would be nice to have an option to use the crop box instead, like:

convert_from_path(..., use_crop_box=False)

Cheers

seems use_cropbox=False, refer to https://pdf2image.readthedocs.io/en/latest/reference.html