I experimented with this last year. I was trying to prototype a client-server system on which graphics were rendered on a central server then compressed and piped to clients.
I played with some wavelet video compression/decompression cards based on the analog devices ADV601 chip (you can google it). It can achieve high compression ratios on grayscale images working on a frame by frame basis (kinda like MJPEG but with wavelets).
After playing with the server a bit (it was a Beowulf cluster:) I wrote a software wavelet codec which I then tried to integrate with MPEG2 interframe compression. This turned out to be very tricky because a lot of the interframe motion vector compression relies on the DCT blocks from the JPEG-style intraframe stage (you've probably seen the obvious 'boxes' of pixels when viewing a very highly compressed JPEG image).
Anyway, the results I was getting (for grayscale) *sound* impressive. 200:1 was possible for most images but only pictures with smooth contrast changes looked any good after decompression. Any sharp edges (e.g. graphical overlays) were completely destroyed at any compression rate over 10:1. Throwing the MPEG interframe stuff into the mix didn't really help much (partly due to the problem outlined above), although I can't say I explored all the possibilities along this route.
After becoming more interested in coding proper parallel apps for Beowulves rather than hacking the MPEG's source I let the project drop. Code available if you'd like a look.
My personal opinion on this fullscreen video with CD-quality sound over 28.8 is that it's complete tosh. It's absolutely impossible to compress that much information into such a small pipe. Unless this guy has discovered something that makes an awful lot of our current mathematical thinking invalid then this claim is nonsense.
Slashdot Mirror
I experimented with this last year. I was trying to prototype a client-server system on which graphics were rendered on a central server then compressed and piped to clients.
:) I wrote a software wavelet codec which I then tried to integrate with MPEG2 interframe compression. This turned out to be very tricky because a lot of the interframe motion vector compression relies on the DCT blocks from the JPEG-style intraframe stage (you've probably seen the obvious 'boxes' of pixels when viewing a very highly compressed JPEG image).
I played with some wavelet video compression/decompression cards based on the analog devices ADV601 chip (you can google it). It can achieve high compression ratios on grayscale images working on a frame by frame basis (kinda like MJPEG but with wavelets).
After playing with the server a bit (it was a Beowulf cluster
Anyway, the results I was getting (for grayscale) *sound* impressive. 200:1 was possible for most images but only pictures with smooth contrast changes looked any good after decompression. Any sharp edges (e.g. graphical overlays) were completely destroyed at any compression rate over 10:1. Throwing the MPEG interframe stuff into the mix didn't really help much (partly due to the problem outlined above), although I can't say I explored all the possibilities along this route.
After becoming more interested in coding proper parallel apps for Beowulves rather than hacking the MPEG's source I let the project drop. Code available if you'd like a look.
My personal opinion on this fullscreen video with CD-quality sound over 28.8 is that it's complete tosh. It's absolutely impossible to compress that much information into such a small pipe. Unless this guy has discovered something that makes an awful lot of our current mathematical thinking invalid then this claim is nonsense.