Classification of analytical images with radial basis function networks and forward selection

Abstract

A novel method for classification of images based on the architecture of radial basis function networks (RBFN) is proposed. The classification algorithm uses the concentration histogram image (CHI) of two images as a density function that can be approximated with a large number of radial basis functions (RBFs). With forward selection, the number of these basis functions can be reduced to a given number of the most important basis functions that represent the cluster centers of the image set. These centers are used for the following classification process based on the distance and weight of the nearest basis function. This novel classification approach is compared to common classification methods and is illustrated with the example of analytical images but not restricted to them.