Color texture analysis and classification: An agent approach based on partially self-avoiding deterministic walks

Recently, we have proposed a novel approach of texture analysis that has overcome most of the state-of-art methods. This method considers independent walkers, with a given memory, leaving from each pixel of an image. Each walker moves to one of its neighboring pixels according to the difference of intensity between these pixels, avoiding returning to recent visited pixels. Each generated trajectory, after a transient time, ends in a cycle of pixels (attractor) from where the walker cannot escape. The transient time (t) and cycle period (p) form a joint probability distribution, which contains image pixel organization characteristics. Here, we have generalized the texture based on the deterministic partially self avoiding walk to analyze and classify colored textures. The proposed method is confronted with other methods, and we show that it overcomes them in color texture classification. © 2010 Springer-Verlag.