Noise-resistant local binary pattern based on random projection

Local binary pattern (LBP) is sensitive to noise. LBP projects local patch to eight-dimension vector by operating subtractions between pixel and its neighborhood. Two adjacent pixel values are generally very close, thus little noise can change their relative magnitude, leading to coding err. Using another projecting approach, random projection, as an alternate, we propose local binary pattern based on random projection (RPLBP), which is much more robust to noise while remaining computationally simple. Experiments on the aligned FERET database show that RPLBP has outstanding robustness to variations in illumination, facing expressions and aging. It's worth mentioning that the proposed RPBLP demonstrates superior noise-resistant performance to LBP and NRLBP.