Image datastream with Flate compression results in "cannot identify image file" error

[bitsgalore]

Example:

        <image>
          <dict>
            <xref>69</xref>
            <width>450</width>
            <height>369</height>
            <bpc>8</bpc>
            <colorspace>DeviceRGB</colorspace>
            <altcolorspace></altcolorspace>
            <filter>FlateDecode</filter>
          </dict>
          <stream>
            <exceptions>
              <exception>cannot identify image file &lt;_io.BytesIO object at 0x7f9795784c70&gt;</exception>
            </exceptions>
          </stream>
        </image>

Example file:

https://www.dbnl.org/tekst/kris001leze09_01/kris001leze09_01.pdf

[bitsgalore]

Image extraction with pdfimages (pdfimages -all kris001leze09_01.pdf ./test-kb/) for this PDF results in a bunch of PNG images.

Extracted stream object to file using:

mutool show -be -o 69.dat kris001leze09_01.pdf 69

Resulting file doesn't look like a PNG, so tried:

file 69.dat

Result:

69.dat: zlib compressed data

So my best guess is that the stream object just contains flate-compressed raw image data, without actually wrapping this in PNG format (which also uses flate compression). This also makes sense, since PDF's FlateDecode filter is used generically for compressing various things, including text.

As a result, the format is not recognised by Pillow, resulting in the exception.

[bitsgalore]

See also:

#31

(of which this is one particular case)

[bitsgalore]

Possible fix would be to restrict stream-level processing to a subset of Filter values for which this actually works. But this doesn't work if the image stream is e.g. an ASCII-85 encoded (see here), in which case PyMuPDF only reports the ASCII85Decode filter value!