This tries to reproduce an issue reported at https://zola.discourse.group/t/jpeg-image-quality/1227
The folder in contains three different images, downloaded from unsplash.com in full resolution (4K+),
then scaled down to 1920x1280 and saved as jpeg with 100% quality in Krita.
These are the input images for the test.
The folder out contains these same images, "resized" to exactly the same size with the image crate.
For each of the three images 10 versions (10%-100% in steps of 10) are generated programmatically.
The folder ref contains the same three images from in, just opened with Krita, and then resaved
as-is with 100% quality. (When doing ls -al on the in and the ref folder one sees that the images
each have significantly diverging sizes between the input and the re-saved versions, indicating that Krita
is indeed performing a re-encode (and not simply saving a verbatim copy).
See src/main.rs for the minimal source code used to generate the images.
Focusing on areas with similar colors, like skies:
- With quality at 100%, the "image" encoder introduces bending and a green tint, compared to no bending and not tint in Gimp with similar settings.
- With quality at 90%, the "image" encoder produced an expected filesize while quality is significantly worse than the Gimp version, which is still nearly visually lossless
- With quality at 70%, quality starts to become compareable to the gimp version, while a slightly greenish tint remains.
There was no difference in quality whether the resize function was used or not. There was no visible difference in quality in more complex high-frequency parts of the image.
