Images not extracted

[MartinThoma]

When trying to extract the images, PyPDF2 didn't capture them.

Environment

$ python -c "import PyPDF2;print(PyPDF2.__version__)"
2.11.0

Code + PDF

Use this PDF: https://arxiv.org/pdf/2201.00151.pdf

>>> from PyPDF2 import PdfReader
>>> reader = PdfReader("2201.00151.pdf")
>>> reader.pages[1].images
[]

Expected were two images.

[MartinThoma]

The missing ones might be inline images:

An inline image object uses a special syntax to express the data for a small image directly within the content stream.

The operators are:

• BI: Begin an inline image object.
• ID: Begin the image data for an inline image object.
• EI: End an inline image object.

[MartinThoma]

Examples:

There are a lot of images with width=1 and height=1:

BI
/IM true
/W 1
/H 1
/BPC 1
ID �
EI Q
q 4.8 0 0 -279.6 368.28 8056.48 cm
BI
/IM true
/W 1
/H 1
/BPC 1
ID �

And some with just small dimensions:

BI
/CS/R124
/W 40
/H 40
/BPC 8
/F/Fl
ID x��S[o�E�ݝݙ����ދw��u6q7N�'�o���8��4��ҖT�UH%Z�����
!!���J ����?��������":��iℋ8�����o>�9�o麨����� G�<⡀ �<����X�����@�� K�(P JE����X��NH��dU@��8�ʲ��������b]#2�r#�Q	������1��Z��l���r��C��r���MxX2�R����	�(��LjrY~,
&�	noԪ��i:�mۊ��T�b�80�1�c��}e�\v�0(Z�f����Y!�Q9���Y{6���LMeհE��,
\^
UE�㹼aQQ�.���'�Z�E�\�HrY5�t���zzi�V	<�sf���:���&�d	�<;I��j�(��`���yaэ��k���^��@���_�l�������MR�(�cn���5'+��{f���q�$����(,���Ndx\2+����v�(Z����b���Գ�3���
�J���K��@��l��B\�+�WN�R�����U��aD?i6[�S��k�{��?�?���[M�� ��/�����N�v#��ԖV��7�+�I��
CUD�9�����l%���XovV��a���j�,T|C����	��(��bP���[k�;ۣ^w�l9
����L?�,���Rg�����Q�KJ�*Ɯ�"H�fsi4�ܹys��F;�,M��{�������/&qm�So�΢�PP�U��S��Gw���rZ�
v�~k��dQb�1��������w�o}������ӧ{��M$HOlʕa�o���G���ܸV_���!����������no��mm]�//d�O�A���S��W_���x�����J\*�tSg���~��������n�/�6��0�u��4�����n�{��g��߸����e������/?~r����|���O�?���v#�u��{��<���õK6�Ӌ�������/��:��5z���
���׳�"�������W�1��Fc������ƫqo
EI

[MartinThoma]

https://arxiv.org/pdf/2201.00214.pdf contains even more missing images

[MartinThoma]

PyMuPDF extracted a couple of images for https://arxiv.org/pdf/2201.00029.pdf whereas PyPDF2 didn't extract any.

Here is how to extract images with PyMuPDF: https://github.com/py-pdf/benchmarks/blob/main/benchmark.py#L185-L198

[stchris]

Not sure if this helps, but I thought I'd bisect this to see if it's a regression. My test script is based on a snippet from above, but goes through all pages:

from PyPDF2 import PdfReader
import sys

reader = PdfReader("2201.00151.pdf")
images = []
for page in reader.pages:
    images.extend(page.images)

if images == []:
    print("bad")
    sys.exit(1)
else:
    print("good")

I went as far back as 85b3e87 where PageObject.images was implemented and it seems to be consistently faulty.

[stchris]

Oh and I seem to have missed the docstring saying this doesn't work with inline images: https://github.com/py-pdf/PyPDF2/blob/0b2b3ec997b81d6bb22c5c834ef5ba9e4f2330e7/PyPDF2/_page.py#L378

[pubpub-zz]

2nd page:

4th image: