Face recognition based on modified LBP

Face recognition in unconstrained, natural conditions still remains a challenging task. As a powerful local descriptor, Local Binary Patterns has shown the advantage of representation and performance. However, it is still affected by robustness and accuracy. In this paper, a novel method is presented to improve the performance of automatic face recognition under uncontrolled conditions. We modify the conventional Local Binary Pattern and use it as a new feature descriptor. Partial Hausdorff Distance is applied as a dissimilarity measurement. Experimental results show that the proposed algorithm outperforms the traditional LBP approach in terms of accuracy rate and robustness. It can reduce the sensitivity caused by illumination variation, pose variation, occlusion etc.