Dominant-Completed Local Binary Pattern for Texture Classification

this paper presents a new approach to extract image features for texture classification. The extracted features are obtained by a dominant-completed modeling of the traditional local binary pattern (LBP) operator, which is robust to image rotation, grey scale changing and insensitive to noise and histogram equalization. The main idea of this texture classification approach is that a dominant center pixel and dominant local difference sign-magnitude transforms (DLDSMT) are used to represent the local region of a texture image. The dominant center pixels represent the gray level of a texture image and they are transformed into a binary code by a global threshold, namely DCLBP_C. The image local differences, by using DLDSMT, are decomposed into two complementary components: the dominant signs and the dominant magnitudes. And they are also transformed into binary codes, namely DCLBP_S and DCLBP_M. By converting DCLBP_S, DCLBP_M, and DCLBP_C features into joint or hybrid distributions, we can obtain our proposed feature. Experimental results reveal that our proposed method outperforms several representative methods.