Wavelet based texture classification with evolutionary clustering networks

This paper deals with the development of an unsupervised pattern classification technique that exploits the searching capability of genetic algorithms for automatic clustering of a given set of feature vectors into an appropriate number of clusters using the Kohonen self organizing map (SOM) technique and adaptive resonant theory network (ART). Since the number of clusters is not known a priori, a modified string representation is used in order to encode the variable number of clusters. The percentage of correct classifications is used as a measure of fitness of a chromosome. In addition, wavelets are used for data pre-processing. The effectiveness of the evolutionary classification scheme is demonstrated for 7 classes of Brodatz texture. The performance with SOM and ART clustering is evaluated and the results show the efficiency of the genetic algorithm in selecting various parameters and clusters in developing an automatic texture classification method.