Phoronix: Google Comes Up With Its Own Image Forms: WebP
After previously open-sourcing the VP8 video codec and coming up with a new container format (WebM), Google set its sights on making a new image format. Google has now publicly announced and released the initial code to the WebP image format. The goal of WebP is to better compress images than PNG and JPEG files commonly used on web-sites while retaining the same image quality...
I thought the size reductions varied from decent to impressive. No visible change in quality;
However, at these sizes you probably couldn't tell the difference between a more slightly compressed jpeg either.
"Images and photos make up about 65% of the bytes transmitted per web page today"
..if you don't count videos, of course, but 65% is just a nice number so we'll keep it there.
"we randomly picked about 1,000,000 images from the web"
..but didn't bother optimizing theim with jpgcrush et al, so we'll just benchmark encoder efficiencies instead of format improvements.
"..and re-encoded them to WebP without perceptibly compromising visual quality"
..and please don't ask how we measure "perceptibly compromised". Hey, when you just move the jpeg slider from 75% to 70% you can get quite a file size reduction, and most of the time I don't perceive any quality difference, either!
"We expect that developers will achieve in practice even better file size reduction with WebP when starting from an uncompressed image."
..please implement this new format in every single app since 1992 based on our unproven expectations.
"We plan to add support for a transparency layer, also known as alpha channel in a future update."
..we would have done it by now, but frankly we aren't sure how to add a useful alpha channel to a lossy compression method, either. Let's markt it as "future work".
"The tables below show scaled JPEG and WebP images side-by-side for comparison."
..we prefer to show you scaled down versions so you won't actually notice any difference in visual quality. Isn't WebP great?
The only real benchmark would be to take uncompressed images, compress them as good as you can with both jpeg and webp, then compare the improvements. Either compress to the same quality and compare sizes, or compress to the same size and compare quality. Any comparisons of visual quality should be done as a double-blind test.
I wonder why google didn't set up a simple script showing you two compressed versions of an image and letting you pick the better one?
They did bother trying to optimize the jpegs and found that optimizing them gave ~14% better compression while using WebP gave ~39% (both compared to the original images). So the improvements of WebP over well compressed jpeg images is ~29%.
Guess which number appears in the blog post? That's right, 39%!
Their method for measuring visual quality was the PSNR. Conveniently, that's exactly the metric WebP is optimizing for. Inconveniently, it's only loosely related to perceived visual quality, because it completely lacks any psychovisual considerations. The link in my last post shows very well why that is a problem.
In other words, WebP can achieve a broken metric 29% more efficient than jpeg. Does it create better looking images at lower filesizes? That one still remains doubtful. Seeing that google purposefully misleads the reader in several aspects of their presentation ("39%", scaled down images, research details remain hidden, no talk about feature parity, ..) I remain sceptical.
This hasn't proven to be technically superior to jpeg yet, and it damn well needs to be if it wants to get the market penetration it needs to survive.