rise and fall and rise of jpeg 2000

[boxerab]

Latest release of grok jpeg 2000 toolkit has full support for new high throughput Part 15 of the standard.
(courtesy of the OpenJPH project)

From informal benchmarks on lossless greyscale encoding, I am seeing ~ 5x increase in
decode speed at the cost of ~8% compression performance.

This HT implementation is multi-threaded but not optimized in other ways, so there is potential
for much faster decode, I think.

https://github.com/GrokImageCompression/grok/releases
https://github.com/aous72/OpenJPH

I am planning on doing some speed/image quality/compression performance comparisons with
AVIF and JPEG XL, the two other open contenders for JPEG successor.

[Scope]

https://medium.com/@scopeburst/mozjp...n-44035c42abe8
I did some visual comparisons of HTJ2K, I don’t think that for now it will be a competitor to more modern image formats (at least for the Web), sometimes it was even worse than MozJpeg.
But it will be interesting to see a more detailed comparison.

[Jyrki Alakuijala]

I am planning on doing some speed/image quality/compression performance comparisons with
AVIF and JPEG XL, the two other open contenders for JPEG successor.

During the JPEG XL standardization process I was surprised how good JPEG 2000 has become. My impression from its early days was that it wasn't working very well (good objective numbers, but images looked bad).

JPEG XL, AVIF, and the best H.265-based image codec (publicly available bpgenc does some strange hanky-panky with default settings) are all far better than JPEG 2000 for the usual image compression needs.

[boxerab]

Scope: just curious about your settings for OpenJPEG in that comparison you did: "-s 1,1 -q x" . OpenJPEG doesn't actually sub-sample here, but just doubles the dimensions of the canvas. Also, the quality layer in DB is set to "x"; what value did you use for the PSNR ?

[Scope]

Don’t remember exactly where I got these settings from (probably from one of the frameworks), but I tried to imitate Netflix's encoding, however I tested with other settings and did not notice noticeable differences in quality (these images were added to the blog with exact settings).

[boxerab]

Scope: thanks, this must be a related blog: https://netflixtechblog.com/avif-for...ng-b1d75675fe4 . Is the benchmark code available ?

[Scope]

Initially, I did my tests to demonstrate that Jpeg is not as bad as in their examples (but over time I began to add other comparisons). The Netflix benchmark (Framework) is on the Github, but I did not create or use the similar ones, everything was encoded manually and with a couple of batch files.