Identification of stochastic textures with multiresolution features and self-organizing maps

An automatic method of clustering and identifying stochastic textures by means of self-organizing maps is presented. The idea is to utilize co-occurrence matrices at different resolution levels and to let the self-organizing process take care of the clustering problem. The labeling is done by identifying the known samples on the map. The unknown samples can be classified by the nearest-neighbor method. The procedure has been tested with natural textures. The results obtained have been promising