Tiff image JPEG compression up to 3x original image disk size

[GiovanniNova]

I have been using this library for a while and although it works extremely well and is much faster than say, OpenCV, I noticed that it has some issues with compression.

When converting an image from JPEG to TIF they can end up with over 3x their original disk size, this woudn't be an issue for 1 or 2 images but we have over 4.000 images.

Sample code:

from PIL import Image

with Image.open('original.jpg') as image:
	image.save('pillow - 50.jpg', optimize=True, compression='jpeg', quality=50)
	image.save('pillow - 50.tif', optimize=True, compression='jpeg', quality=50)
	
	image.save('pillow - 75.jpg', optimize=True, compression='jpeg', quality=75)	
	image.save('pillow - 75.tif', optimize=True, compression='jpeg', quality=75)

Original 89%: 3.352 kb
Photoshop 9/12 (75%): 3.417 kb
Pillow 50% .JPG: 1.175 kb
Pillow 50% .TIF: 3.828 kb
Pillow 75% .JPG: 2.320 kb
Pillow 75% .TIF: 7.206 kb

original.zip
result.zip

• Windows 10 x64
• Python 3.9.6
• Pillow 9.4.0
• LibTiff 3.8.2-1 (features.check('libtiff') returns True)

[radarhere]

Setting aside quality and optimize, I tried comparing the results with ImageMagick's convert command.

convert original.jpg -compress JPEG magick.tif gave 7,203kb.

The Pillow code

from PIL import Image
with Image.open('original.jpg') as image:
    image.save('pillow.tif', compression='jpeg')

gave 7,205kb.

So we're very similar to ImageMagick's output.

Also, just as an aside, compression is not a keyword argument used when saving JPEG images - https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html#jpeg-saving

[GiovanniNova]

Also, just as an aside, compression is not a keyword argument used when saving JPEG images -

It was a small mistake since copy + pasted the TIF save lines and forgot to remove the compression argument

So we're very similar to ImageMagick's output.

So, is this a won't fix kind of situation? Is Pillow a wrapper around ImageMagick? If so, should I bring this issue to ImageMagick's repo instead?

[cgohlke]

The size difference likely comes from Photoshop compressing JPEG in YCBCR colorspace inside TIFF, while Pillow and ImageMagick use RGB colorspace. Try convert original.jpg -compress JPEG -colorspace YCbCr magick.tif

[GiovanniNova]

Confirmed working by using Image.convert('YCbCr'), which produces a TIF file with a very small difference of 400KB from their JPEG counterparts, thank you!

[Yay295]

I think it would also technically be possible to embed a JPEG into a TIFF without recompressing, but I think that would be out of scope for Pillow.

edit: I found some experimental C# code to do exactly this. https://stackoverflow.com/a/14877473/3878168

[Yay295]

I found an open issue on the libtiff GitLab for this same situation: https://gitlab.com/libtiff/libtiff/-/issues/106

[radarhere]

Is Pillow a wrapper around ImageMagick?

Not at all! I was just saying that two independent tools have similar results, so this is likely not as simple as "Pillow has a bug".

Confirmed working by using Image.convert('YCbCr'), which produces a TIF file with a very small difference of 400KB from their JPEG counterparts, thank you!

Is this issue resolved then?

[Yay295]

I wrote a function that can directly embed a JPEG file into a TIFF file: #7006

Using your original.jpg I converted it to TIFF with only 204 bytes of overhead.

from jpeg_to_tiff import jpeg_to_tiff
with open('embedded.tif', 'wb') as tiff_file:
    tiff_data = jpeg_to_tiff('original.jpg')
    tiff_file.write(tiff_data)

[cgohlke]

I wrote a function that can directly embed a JPEG file into a TIFF file:

Nice. The YCbCrSubSampling tag needs to be specified for YCBCR images unless they are YUV420.

[GiovanniNova]

Is this issue resolved then?

Thank you, it has been solved.