I want to do a project where I sort images by color, more or less pixel by pixel. If a pixel in one picture is the same color as a pixel in the same spot in another picture, that would be a score of 1. If it's a similar color, it's a lower score. If there is the same or a similar color nearby, that gets a different score. No nearby matches or similarities gets 0. Images with hight relative scores would go next to each other, to try to get a progression of images that have a high similarity between each other in a sequence.
Obviously, scoring a large number of these images would be extremely computationally intensive, so I plan to not calculate all of them, but do a genetic algorithm to try out different progressions and get one that's pretty good, even if not the best. The big problem, though, is the comparisons. There are Perl libraries for dealing with images and I can write some code to do it, but it's fairly work intensive. I think some program that does this must already exist, but I haven't found anything. Anybody out there heard fo anything that will do this? It doesn't have to be 100% fully featured. Just knowing there's the same color in the same spot is good enough. I could find similar colors vs the same color by first comparing the originals and then comparing reduced color versions. If the originals match, the score is 1. If the originals don't match, but the reduced color versions do, then the score is 0.5. No match at all is 0. For example.
In unrelated news, I've just upgraded this blog to the new google blog beta. It seems to have support for tags built-in. We'll see.
1 comment:
My friend Ronnie Cramer used a freeware program (whose name escapes me) to convert the colors of paperback novels he scanned into synth sounds.
Post a Comment