A fuzzy ARTMAP based classification technique of natural textures

Describes an approach to the classification of textured gray-scale images using a technique based on image filtering, the fractal dimension (FD) and a fuzzy ARTMAP neural network (FAMNN). 12 FD features are computed based on 12 filtered versions of the original image using directional Gabor filters. Features are computed in a window and mapped to the central pixel of this window. We implemented a variation of the testing phase of a fuzzy ARTMAP that exhibited a performance that was superior to the standard fuzzy ARTMAP and the 1-nearest neighbor (1-NN) method in the presence of noise. Training was performed using patterns that were extracted from 20 different textures. The classification performance is also studied with respect to a testing set. Segmentation results are also presented to illustrate that the classification algorithm and its specified parameters are adequate so that more than one texture can be identified in the same image.