PngStream does not decode EXIF data from "Raw profile" tags

[alexdw-mgzi]

What did you do?

I attempted to read the meta-data from one of the test images from Exiv2 ("ReaganSmallPng.png"). This file contains EXIF and IPTC meta-data stored under compressed "Raw profile type" tags, typical of ImageMagick (and possibly other software).

What did you expect to happen?

• The Image's info["exif"] contains the EXIF data, and getexif() returns an object containing some or all of the meta-data, or
• The Image's info["Raw profile type exif"] contains the EXIF data, which can be manually fed back into info["exif"]

What actually happened?

The Image's info["exif"] was empty, and getexif() returned no metadata. The "Raw profile type exif" was read, but was not suitable to be used as info["exif"]. (Attempting this produces a SyntaxError in the TiffImagePlugin, which I haven't investigated further.) I tried modifying PngStream.chunk_zTXt in "PngImagePlugin.py" to skip decoding the bytes as "latin-1", but I still got the same SyntaxError.

Found Exif tags: 0
Traceback (most recent call last):
  File "pil_err.py", line 9, in <module>
    exif = reagan.getexif()
  File "/usr/lib64/python3.6/site-packages/PIL/PngImagePlugin.py", line 704, in getexif
    return ImageFile.ImageFile.getexif(self)
  File "/usr/lib64/python3.6/site-packages/PIL/Image.py", line 1275, in getexif
    self._exif.load(self.info.get("exif"))
  File "/usr/lib64/python3.6/site-packages/PIL/Image.py", line 3237, in load
    self._info = TiffImagePlugin.ImageFileDirectory_v1(self.head)
  File "/usr/lib64/python3.6/site-packages/PIL/TiffImagePlugin.py", line 900, in __init__
    super().__init__(*args, **kwargs)
  File "/usr/lib64/python3.6/site-packages/PIL/TiffImagePlugin.py", line 466, in __init__
    raise SyntaxError("not a TIFF file (header %r not valid)" % ifh)
SyntaxError: not a TIFF file (header b'    8526' not valid)

What are your OS, Python and Pillow versions?

• OS: Gentoo Base System release 2.6, Linux kernel 4.19.97-gentoo x86_64
• Python: Python 3.6.10
• Pillow: Pillow 6.2.2 (from Gentoo), Pillow 7.0.0 (from Gentoo)

import PIL.Image

reagan = PIL.Image.open("ReaganSmallPng.png")
exif = reagan.getexif()
print("Found Exif tags:", len(exif.keys()))

reagan.info["exif"] = reagan.info['Raw profile type exif'].encode('latin-1')
exif = reagan.getexif()
print("Found Exif tags:", len(exif.keys()))

Image causing the issue

Additional information

ImageMagick's PNG encoder:
https://github.com/ImageMagick/ImageMagick/blob/master/coders/png.c

ExifTool's guide to PNG Tags:
https://exiftool.org/TagNames/PNG.html

[radarhere]

I found https://sourceforge.net/p/png-mng/mailman/png-mng-misc/?viewmonth=201612&viewday=31 which pointed out that the data is hexencoded. So I find that this gives me 55 EXIF keys.

from PIL import Image

reagan = Image.open("ReaganSmallPng.png")
exif = reagan.getexif()
print("Found Exif tags:", len(exif.keys()))

reagan.info["exif"] = bytes.fromhex("".join(reagan.info['Raw profile type exif'].split("\n")[3:]))
exif = reagan.getexif()
print("Found Exif tags:", len(exif.keys()))

[alexdw-mgzi]

Thanks, that's what I was missing!

With that as a "work-around", I think that this issue could be classified as a feature-request rather than a bug.

[radarhere]

Looking into this, ImageMagick implemented "Raw profile type exif" before Exif became part of the PNG standard - https://ask.xiaolee.net/questions/1122297. It doesn't write both - https://github.com/ImageMagick/ImageMagick/blob/868aad754ee599eb7153b84d610f2ecdf7b339f6/coders/png.c#L11068-L11071. So it seems reasonable to alter PngImagePlugin to read this as well.

However, this means it is slightly harder to generate an image with a "Raw profile type exif", since ImageMagick hasn't written it in two years. Would you happen to have an image that could be included in the test suite under the Pillow license?

[radarhere]

I managed to create a file, and have created PR #4471 to resolve this.