Color Texture Classification Using Local Texture Features

Abstract

This paper proposes a new approach to extract the features of a color texture image for the purpose of texture classification. Four local texture feature sets are concerned. Opponent Color Dominant Neighborhood Structure (OCDNS) and Angular Color Dominant Neighborhood Structure (ACDNS) are the newest local texture feature set that has been used for color texture image classification. Initially the RGB color texture is converted into YCbCr color components. OCDNS is obtained using each pair of different color channels. The intensity resemblance between a given image pixel and its surrounding neighbors are obtained for different color channels. Similarly ACDNS is another feature set obtained after taking an arc tangent between each pair of different color channels. Opponent Color Local Binary Pattern (OCLBP) and Angular Color Local Binary Pattern (ACLBP) features are also extracted and compared with that of above feature sets. In addition to that Local Binary Pattern and Dominant Neighborhood Structure features are also extracted and different combinations were made to improve the classification rate. Also the proposed feature sets are robust to noise. The texture classification process is carried out with the robust SVM classifier. The experimental results on the CUReT database shows that the proposed method is an efficient method whose classification rate is higher when compared with the other methods.