Sparse Representation Algorithm Based on Block LBP and Adaptive Weighting

To solve the problem of low recognition rate caused by occlusion, illumination and pose change in face image, a sparse representation algorithm based on block LBP and adaptive weighting (SBLAW) is proposed. Firstly, the face image is divided into equal blocks and LBP value of the sub-blocks is calculated, then the histogram of each sub-block is counted and weighted adaptively. Secondly, the image is divided into overlapping blocks, and the statistical histograms of the sub-blocks in the overlapping blocks is concatenated as the feature vector. Finally, the sparse representation classification method is used for face recognition. The experiments on the ORL, AR and FERET face databases show that the proposed method is robust to illumination and other factors, and the recognition rate is improved.