I'm working now on a project where we need to detect and recognize different logos in video files. Adidas logo were chosen to be my laboratory rat. I trained a cascade and it gave good results. But only for black logo, so i trained another cascade for same shape but the color of the logo was white now, and it also gave good results. Now what do i do with all those colored logos ? I tried contour based algorithms cause they color indifferent, but they did not give good results, especially for logos like Adidas where there several parts : 
and i should combine all parts together somehow. My question is how can i make all bright logos to became white, and dark logos to become black, which manipulations can cause such effect, and if it is possible would cascades still work on such images ? Or there might be some combination of algorithms (cascade + contour) , or there some other approach exists that fits my problem ? Thanks in advance.
Performance not so important.
