A statistical design of experiments approach for texture description

Abstract

In this paper a method of texture description is proposed based on a closed set of orthogonal effects due to spatial variation of the gray levels. The statistical design of the experiments paradigm has been used to suitably choose and then measure the significance of orthogonal effects towards texture. The measure of significance is encoded as a number called pronum. The frequency of occurrences of pronums called prospectrum is used as the global descriptor of textured images. The proposed technique is applied for classification of various natural textured images resulting an average of correct classification up to 92%.