PIL does not recognize dpi of some jpeg images

[dziket]

It seems that for some jpeg images Pillow does not report correct DPI.

For instance, I created with Gimp a PNG file of size 1x1 pixels with resolution of 123 dpi (test2.png) and then I converted it with imagemagick to test2.jpg. The command was:

$ convert --version
Version: ImageMagick 6.9.10-23 Q16 x86_64 20190101 https://imagemagick.org
Copyright: © 1999-2019 ImageMagick Studio LLC
License: https://imagemagick.org/script/license.php
Features: Cipher DPC Modules OpenMP 
Delegates (built-in): bzlib djvu fftw fontconfig freetype jbig jng jpeg lcms lqr ltdl lzma openexr pangocairo png tiff webp wmf x xml zlib
$ convert test2.png test2.jpg 

I attach both files (zipped, as images.zip). When I use Pillow to read dpi of test2.jpg, this is what I get:

$ ipython3 

In [1]: from PIL import Image                                                                                                                                 

In [2]: print(Image.__version__)                                                                                                                              
10.4.0

In [3]: Image.open('test2.jpg').info                                                                                                                          
Out[3]: {'jfif': 257, 'jfif_version': (1, 1), 'jfif_unit': 2, 'jfif_density': (48, 48)}

I see that jfif_density and jfif_unit pair is correct (48 px/cm is 123 px/in), but I think that this info structure should also contain 'dpi' entry - and it does not contain it. For instance, when I create a similar jpeg file directly from Gimp (test3.jpg, also attached), it works as I want:

$ ipython3 
/usr/lib/python3/dist-packages/IPython/core/interactiveshell.py:935: UserWarning: Attempting to work in a virtualenv. If you encounter problems, please install IPython inside the virtualenv.
  warn("Attempting to work in a virtualenv. If you encounter problems, please "
Python 3.8.10 (default, Nov  7 2024, 13:10:47) 
Type 'copyright', 'credits' or 'license' for more information
IPython 7.13.0 -- An enhanced Interactive Python. Type '?' for help.

In [1]: from PIL import Image                                                                                                                                 

In [2]: print(Image.__version__)                                                                                                                              
10.4.0

In [3]: Image.open('test3.jpg').info                                                                                                                          
Out[3]: 
{'jfif': 257,
 'jfif_version': (1, 1),
 'dpi': (123, 123),
 'jfif_unit': 1,
 'jfif_density': (123, 123),
 'exif': b'Exif\x00\x00II*\x00\x08\x00\x00\x00\x07\x00\x0e\x01\x02\x00\x12\x00\x00\x00b\x00\x00\x00\x1a\x01\x05\x00\x01\x00\x00\x00t\x00\x00\x00\x1b\x01\x05\x00\x01\x00\x00\x00|\x00\x00\x00(\x01\x03\x00\x01\x00\x00\x00\x02\x00\x00\x001\x01\x02\x00\r\x00\x00\x00\x84\x00\x00\x002\x01\x02\x00\x14\x00\x00\x00\x92\x00\x00\x00i\x87\x04\x00\x01\x00\x00\x00\xa6\x00\x00\x00\x00\x00\x00\x00Created with GIMP\x00{\x00\x00\x00\x01\x00\x00\x00{\x00\x00\x00\x01\x00\x00\x00GIMP 2.10.18\x00\x002024:12:28 07:13:34\x00\x02\x00\x86\x92\x07\x00\x19\x00\x00\x00\xc4\x00\x00\x00\x01\xa0\x03\x00\x01\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00Created with GIMP\x00',
 'progressive': 1,
 'progression': 1}

[radarhere]

Looking at the code, Pillow does use the JFIF density to set info["dpi"] - if the JFIF unit is 1, meaning that the JFIF density is per inch.

Pillow/src/PIL/JpegImagePlugin.py

Lines 91 to 94 in e8dad19

	if jfif_unit == 1:
	self.info["dpi"] = jfif_density
	self.info["jfif_unit"] = jfif_unit
	self.info["jfif_density"] = jfif_density

So your request is that when the JFIF unit is 2, meaning that the JFIF density is per cm, then info["dpi"] should be inferred from that? That seems reasonable. I've created #8633

[dziket]

Yes, you summarized the problem better than me - the problem is "what should be the value of DPI if JFIF unit is 2".
The whole thing have started when I noticed today that img2pdf (which uses Pillow) do not support correctly files in which JFIF unit is 2 (see: josch/img2pdf#40 ). So now some code needs to support this scenario - either img2pdf or Pillow itself. And probably it's better if Pillow takes care of it.