3 ms·
Thanks for this comment. I hate how many analyses use PSNR to represent “quality” (especially academic papers; I don’t understand why academic CG/CV people have
by jacobolus 12y ago
Thanks for this comment. I hate how many analyses use PSNR to represent “quality” (especially academic papers; I don’t understand why academic CG/CV people have such a love for nonsense metrics).
I think you’re being a bit too hard on JPEG 2000. If you try compressing images to a very very tiny size, I find that the regular old JPEG encoders I’ve tried produce much more objectionable artifacts than the JPEG 2000 encoders I’ve tried, at the same filesize. I haven’t done any very comprehensive study though; do you have a link to a study showing JPEG 2000 performing worse? What type of images and what level of compression was that? You’re right that the JPEG 2000 encoders and decoders have abysmal performance though, making them impractical in many (or even most) use cases.
The big use case I can think of for JPEG 2000 is archival of very high resolution images where disk space is limited/expensive and encoding/decoding speed doesn’t matter much.
I personally thought Google’s WebP was very underwhelming on all the stuff I threw at it 2–3 years ago when I did some ad-hoc personal tests. Basically pretty similar in filesize vs. subjective quality to regular old JPEGs for the images I tested (and noticeably worse than JPEG 2000 at the same file sizes), except brand new and not supported by anyone but Google.
[But note my little hour of ad-hoc testing based on personal judgments of a small handful of images was certainly far from comprehensive or authoritative, and it was a few years ago, so YMMV.]
Seemed to me like the advantages were more political (i.e. giving Google control and helping push support for their video codec and their fight against MPEG license fees) than technical. And it didn’t help that their publicly published comparisons and marketing material were full of obviously bogus tests. Felt like snake-oil sales.
Have they improved the WebP codec since then, or is there a better analysis somewhere of when WebP is a better choice than JPEG?