Optimum window-size computation for moment based texture segmentation

Abstract

The quality of texture segmentation depends on extracted features. Most statistical feature extraction techniques require an optimum region size, called a window, to capture a better texture feature. The literature shows that window-size selection is primarily done by visual inspection based on experience or trial and error. The paper investigates the issue and attempts to formulate a framework based on the established technique of Fourier analysis to automate the optimum window size computation and feature weight selection. Fourier data in polar form has been used for computing the optimum window size and then for generation of the weighted feature space. Clustering using competitive neural networks when applied to moment features extracted using an optimized window shows good results