Texture Classification using Advanced Texton Texture Matrix

Abstract

This paper derives a new framework for texture classification by combining the features of textons with two types of patterns namely symmetric and triangular patterns and with texture features derived from gray level co-occurrence matrix (GLCM) features. The textons derived in this paper are different from the traditional textons and the derived textons are named as advanced textons. This research derived a total of ten textons shapes on a 2x2 grid with two and four identical pixels. This paper initially transforms the given texture image into an advanced texton index (ATI) image. In ATI image each location represents a local shape. This paper then derives symmetric advanced texton unit (SATU) and triangular advanced texton Unit (TATU) on the 3x3 window of the ATI image. A new matrix called advanced texton texture matrix (ATTM) is derived based on the relative frequencies of SATU and TATU codes.  The GLCM features are derived on the ATTM and these features are used as a feature vector. The classification is performed using various machine learning classifiers on the popular texture databases with different types of natural images on the proposed ATTM. The results are compared with state art of existing methods and the results indicate the efficacy of the proposed method over the rest of the methods.