Last week, Google released to open source JPEG encoder that can create images file sizes 35% smaller than currently available methods. Called Guetzli, which means cookie in Swiss German, it seems new possibilities for the format that’s been around since the late 1980s!
Using a different method than other optimizers, Guetzli doesn’t have a set of rules or laws in a race to the lowest file size when encoding images.
Instead, Guetzli relies on a tool that measures differences between images the way the human eye detects differences called butteraugli.
The results on images appear that Guetzli walks a fine line by not applying general compression metric that some artifacts as other methods by focusing on areas of barely noticeable perceptual differences between images.
Guetzli on Images
I ran the CSS Dev Conf speaker photo images to test out the compression. and found that I saved almost 200kb.
This amount of file savings did not apply to all my images, but enough that there was enough for me to go through all my JPEGs on the site!
I even processed a line art image to see how Guetzli would handle the compression and it saved almost 70kb, with barely any noticeable changes.
I wanted to see, if anything, Google’s encoder did anything to that image, I threw both images into the Kaleidoscope app and put them under the Difference view.
Quite astonishing how much work Guetzli did to the image while my eye barely noticed any differences. In fact, I thought the subtle changes enhanced the image!
Future of Guetzli
At the time of my writing, I could not do batch processing, and I did not even try to hook it up to an automated build process since it was slow optimization process.
Hoping to at least be able to do batch processing on the command line soon — something I noticed others were having troubles doing — , but also it does take time for optimization method to do its magic.
Announced last week, I think it’s safe to say this is only the start of Guetzli. I bet shortly the speed will increase, and Guetzli is going to be part of the build processes sooner than later.
Originally published at Christopher Schmitt.