Hyper-Laplacian Regularized Low-Rank Collaborative Representation Classification

Abstract

Face recognition is an important branch of computer vision. Domestic and foreign scholars have proposed many algorithms to improve the face recognition rate. However, when the training sample and the test sample are exposed to light, occlusion or contamination, the performance of the proposed algorithm will decrease. The recently proposed low-rank constrained collaborative representation classification algorithm (LCRC) has been proven to have superior performance in face recognition. The model is a global clustering method that can effectively recover the global subspace structures of the data, but does not consider the local geometric manifold structures of the original data. This will cause it to break the manifold structures of the original data while restoring the data, thereby losing the local geometric information of the recovered data. For the flaws of the algorithm, this paper proposes a hyper-Laplacian regularized low-rank collaborative representation classification (HLCRC). The hyper-Laplacian regularizer is introduced into the low-rank collaborative representation model to maintain the multivariate geometric manifold structures between data. Experiments on public face database show that the proposed algorithm is superior to many existing algorithms in face recognition rate.