Texture classification using dominant gradient descriptor

Abstract

Texture classification is an important part of many object recognition algorithms. In this paper, a new approach to texture classification is proposed. Recently, local binary pattern (LBP) has been widely used in texture classification. In conventional LBP, directional statistical features and color information are not considered. To extract color information of textures, we have used color LBP. Also, to consider directional statistical features, we proposed the concept of histogram of dominant gradient (HoDG). In HoDG, the image is divided into blocks. Then the dominant gradient orientation of each block of image is extracted. Histogram of dominant gradients of blocks is used to describe edges and orientations of the texture image. By coupling the color LBP with HoDG, a new rotation invariant texture classification method is presented. Experimental results on the CUReT database show that our proposed method is superior to comparable algorithms.