And it continues. As I perused the web, I found myself spiralling into some
jpg research and found some docs I decided to quote into some tweets.
So happens that at the same time, the Google Research blog (crossposted in their open source blog as well) posted an announcement regarding a new JPEG encoder: Guetzli. The
jpg is around twice as popular as the 2nd most widely used web format, the
png (which is squarely in the sight of the
webp btw, but more on that soon). As such, at nearly 50% use, any improvements in the
jpg codec holds potential to significantly affect the web, and have positive effects for the user experience.
Guetzli is touting 35% savings over currently available methods. On the practicality side of things, Google is bringing improvements to an existing format, which if you might recall, was Mozilla's argument for introducing Mozjpeg before actually adopting the then new
Seemingly, Guetzli focuses it's improvement efforts in the quantization part of the codec playing with the quality loss, or as they could put it, "[striking] a balance between minimal loss and file size by employing a search algorithm that tries to overcome the difference between the psychovisual modeling of
jpg's format, and Guetzli’s psychovisual model, which approximates color perception and visual masking in a more thorough and detailed way than what is achievable by simpler color transforms and the discrete cosine transform."
However, a very well known challenge with compression is the very hurdle they also mention: time. The research is highlighting the fact that the process is a touch more labour intensive than ideal, with a recent tweet corroborating said fact, but with still positive results around removal of artifacts. But in a stronger position and statement, engineer Nick Doyle added the following:
So far however, research has _apparently_ [proven positive](https://arxiv.org/abs/1703.04416) in the use of the new codec. Best news we can all agree on though, might be the announcement that **ImageOptim** would have [Guetzli](https://github.com/google/guetzli/) in their next release.
@HenriHelvetica For q90+ only, poor on lossy image input, no progressive, ~100x encode CPU, and giant giant memory footprint? No thanks.— Nick Doyle (@njdoyle) March 17, 2017
Nonetheless, make your way to the ==[Google Research Blog post](https://research.googleblog.com/2017/03/announcing-guetzli-new-open-source-jpeg.html)== to see results from the research studies, the details and how you might be able to contribute.
Guetzli will be in the next ImageOptim version. https://t.co/Fwlg2Tbrmt— Kornel (@kornelski) March 16, 2017