Computing Negentropy Based Signatures for Texture Recognition

Abstract

The proposed method aims to provide a new tool for texture recognition. For this purpose, a set of texture samples are decomposed by using the FastICA algorithm and characterized by a negentropy based signature. In order to do recognition, the texture signatures are compared by means of Minkowski distance. The recognition rates, computed for a set of 320 texture samples, show a medium recognition accuracy and the method may be further improved.