Local occluded face recognition based on HOG-LBP and sparse representation

Face recognition system in real environment is easily affected by posture, illumination, shielding and other factors, which makes the recognition effect of traditional face recognition algorithm not ideal. Because most of the current face recognition algorithms do not take into account the fact that there may be occlusion in the actual image, the result of face recognition is wrong. In view of the presence of face occlusion, this paper proposes a face recognition method of local occlusion based on the feature and sparse representation of block-oriented gradient histogram (HOG) and local binary mode (LBP). First, the algorithm segmented the face image and extracted HOG and LBP features from each block to obtain the HOG-LBP joint feature of each block of image, which was projected into the feature subspace by PCA to establish an occlusion dictionary, and then classified and identified by sparse representation reconstruction residual. Compared with the traditional algorithms on ORL and AR face data sets, the proposed algorithm is proved to have better recognition rate and robustness.