Hey, so we were discussing some recent work that used a genetic algorithm to achieve a very interesting, very new way of compressing images by decomposing them into a handful of overlapping polygons. The techniqute brings the prospect for high compression ratios and image artifacts that are more art than nuisance. Take a look here: http://forums.nvidia.com/index.php?s=&…st&p=479761
However, the cost is CPU time. Yet, with an acceleration of one or two orders of magnitude, it might become practical.