First page Back Continue Last page Overview Text


In this table we have a part of the stereo evaluation table that I’ve showed you in the very beginning of my talk. I show only the top 7 algorithms now. Only the global algorithms perform better than our comp.wind.alg., below it are 13 other algl.,some of which are global, some are local. For these images, the running time is from 8 to 22 seconds. For the high texture images, like MAP, our algorithm is better or just slightly worse than most global methods. Compact window is faster than the global methods, so for high texture images it makes more sense to use it.
Again out of the top 6, 5 global algorithms rely on our graph algorithms based global approach, so it’s a good time now to switch to the global approach to stereo correspondence.