A new encoding scheme of LBP based on maximum run length of state “1” for texture classification

As a simple and efficient local feature descriptor, local binary pattern (LBP) is mainly made up of two steps: extraction step and encoding step. In the extraction step, a local region is denoted by a difference vector between the center pixel and its neighbors. In the encoding step, the corresponding binary bit-string of the difference vector is encoded for the following texture classification. Though encoding step plays a vital role in the whole process of LBP, two current widely used encoding schemes of LBPriu2 and LBC still have some limitations. Different from these two current encoding schemes, in this paper, we propose a new LBP encoding scheme based on the maximum run length of state “1” (LBPmr1) in a binary bit-string. The maximum run length of state “1” reflects the most important part of the binary bit-string structure and it is used as the LBP code of a binary bit-string for the first time. Experimental results on four representative texture databases of Outex, UIUC, CUReT and UMD show that the proposed LBPmr1 achieves better classification accuracy compared with other related LBP encoding schemes.