Linked by Thom Holwerda on Thu 30th Sep 2010 23:04 UTC
Google A few months ago, Google open sourced the VP8 video codec as part of the WebM video project, to create a truly Free/free unencumbered video format for the web as an answer to the non-Free/free patent-encumbered H264 format. Today, Google launched a new image format for the web, WebP, which aims to significantly reduce the file size of photos and images on the web.
Thread beginning with comment 443389
To read all comments associated with this story, please click here.
Member since:

commenting about how you can get similar compression rates from JPEG.

Google took a random sampling of images from the web, then re-encoded all of them as JPEG, JPEG2K, and their new WebP format, all targeting the same PSNR rate of 40. Google's format was significantly better, especially at the smaller images.

Now, that doesn't mean this format is an automatic win. How important is another 20% compression compared to the problems associated with supporting a new format? How CPU intensive is the decoding/encoding process? etc. A 3rd party test/comparison would also be good, just to double check that google didn't forget anything important. One good thing is that most of the format is defined in VP8 already, so browsers will already have most of the code in them, it should be just a matter of hooking it up to the correct places to display the images.

Another thing to be aware of, as dark shikari has pointed out VP8 is optimized for PSNR. He shows that that's bad for quality, but it also might have twisted the results of Google's test a bit since they were targeting a constant PSNR of 40 in all 3 formats.

Edited 2010-10-01 08:56 UTC

Reply Score: 5

Timmmm Member since:

Yeah but why on earth did they use JPEGs as the input, and optimise for PSNR?

Also in their comparison gallery all the images are so loosely compressed that there are no differences at all. Totally useless.

Edit: Actually I checked by subtracting the images. They are exactly identical; they must have messed up somewhere.

Edit 2: Ahem, sorry they aren't exactly identical - the differences are nearly all less than 5 though, which is basically imperceptible. If you plot the difference image it looks pure black.

Edited 2010-10-01 13:24 UTC

Reply Parent Score: 2

sakeniwefu Member since:

This guy does a better job. He fucked the JPEGs up at first, but if he has got it right this time then it would match Google's graph and WebP would indeed look better for similarly sized files.

Reply Parent Score: 2