Independent Component Analysis (ICA) for texture classification

Abstract

This paper presents a texture classification algorithm using independent component analysis (ICA). ICA is used for creating basis functions or basis images bank. These basis functions are used in texture classification because they are able to capture the inherent properties of textured images. These properties enable us to use the ICA bank to generate feature vectors for effective texture classification. These feature vectors are used first for training and then for testing the classifier. The experimental setup consists of texture images from the Brodatz Album and a combination of some images therein. Experimental results for both two and multiple class texture have shown that the proposed algorithm which uses ICA has an encouraging performance. With ICA, a large classification improvement was observed. It obtains an average of just 2.85% misclassified pixels compared with 10.26% misclassified pixels by other methods.