jbig2 extractor does not seem to be working correctly

[fgregg]

The jb2 file that is extracted from a pdf with jbig2 encoded images does not seem correct. This includes the sample file

> pd2txt.py --output-dir=. samples/contrib/pdf-with-jbig2.pdf
> ls *.jb2
XIPLAYER0.jb2
> jbig2dec -t pbm XIPLAYER0.jb2
jbig2dec WARNING OOB obtained when decoding symbol instance T coordinate (segment 0x03)
jbig2dec WARNING failed to decode text region image data (segment 0x03)
jbig2dec WARNING failed to decode; treating as end of file (segment 0x03)
> pnmtopng XIPLAYER0.pbm > XIPLAYER0.png

This produces a blank PNG file: XIPLAYER0.png

The original pdf file is okay, and we can extract the image using pdfimages.

> pdfimages -jbig2 samples/contrib/pdf-with-jbig2.pdf XIPLAYER0
> ls XIPLAYER0-000.*
XIPLAYER0-000.jb2e  XIPLAYER0-000.jb2g
> jbig2dec XIPLAYER0-000.jb2g XIPLAYER0-000.jb2e -t pnm
> pnmtopng XIPLAYER0-000.pbm > XIPLAYER0-000.png

This problem is not due to a mistranslation of @side2k's original PR.

I checked out their original Python 2.7 branch, and the jb2 file that they produce is exactly the same as what's on the HEAD of pdfminer.six's develop branch.

[fgregg]

unfortunately, pdfimages (as part of poppler) is licensed under GPL 3, so we cannot read and translate their code and maintain this projects's MIT license

[fgregg]

comparing the output of pdfimages, it looks like XIPLAYER0-000.jb2e and XIPLAYER0-000.jb2g are direct extractions from the pdf. This would be simple to reproduce.

For jbig2 images, we would produce two files, a jb2g and jb2e. Then a user could use jbig2dec to convert these files, together, into pbm or png file.

Obviously, it would be better if pdfminer could output a single jb2 file, but I don't know how to construct a valid file.

It seems better to output two files that can be used, versus one file that is invalid, as is the status quo.

Thoughts @pietermarsman?

[fgregg]

i ended up figuring out how to put everything together in a single jb2 file, and have submitted a pull request. #653

[pietermarsman]

@fgregg Thanks for figuring this out! Will review / merge the PR.