Local Block-Difference Pattern for Use in Gait-Based Gender Classification

In this paper, a novel local texture descriptor termed as Local Block-Difference Pattern (LBDP) is proposed. In conventional LBP, sensitive to intensity change problem will drastically affect the performance due to its simple pixel value comparison mechanism. Different from LBP, the proposed LBDP describes the local textures from a pixel to a block for decreasing the impacts resulting from intensity change. Based on the proposed LBDP, the tolerance to the intensity change is exaggerated because of the expanding of encoding range. The effectiveness of the proposed LBDP is practically demonstrated in the application of gait-based gender classification. In the experiments, CASIA dataset B is adopted and the experimental results demonstrate that the proposed LBDP outperforms the other LBP-based descriptors.