CMYK/Indexed colorspace image extraction

[lambdalemon]

If you are ok with external dependencies it's quite easy to add. For reference I have a PR on pdfminer.six pdfminer/pdfminer.six#1130 that covers most cases (missing weird colorspaces like Lab/Separation/DeviceN and Indexed jpeg/jb2 I think, though tbh I've never seen these cases)

[lambdalemon]

On a related note, in my opinion, the smask of an image should be extracted as a separate image file. If a user wants an image with alpha they should do the compositing themselves.
pypdf tries to add an alpha channel to the extracted image, but this only works if the smask has a different size.

[dhdaines]

On a related note, in my opinion, the smask of an image should be extracted as a separate image file. If a user wants an image with alpha they should do the compositing themselves. pypdf tries to add an alpha channel to the extracted image, but this only works if the smask has a different size.

Yes and also the mask, and the alternates. I don't want to add any dependencies, not even Pillow, so I won't try to do any compositing, the files should just be output (or the data accessible to library users) in a way that makes it easy to do...

[dhdaines]

So, to conclude, I don't want to add CMYK (since it requires a dependency), but indexed colorspaces, yes.

We also need to fix JBIG2 file output which definitely doesn't work at the moment (totally unsure about JPEG2000, have never seen a PDF that has one, but there might be one in the pdf.js test suite).

[dhdaines]

On a related note, in my opinion, the smask of an image should be extracted as a separate image file. If a user wants an image with alpha they should do the compositing themselves.

The best option is always to use a real PDF renderer, since the PDF imaging model is fairly complex. People often want to "extract images" from PDFs without understanding what that actually means ;-)

In any case smask extraction is now there since #135

[dhdaines]

Also, JPEG2000 works just fine by simply dumping out the stream data. Evince can't open it but GIMP has no problem.

JBIG2 is definitely broken, but it's broken in pdfminer.six too. Probably we just need to write the appropriate header (see pdfminer/pdfminer.six#1142)