Robust Image Recognition Algorithm Based On Low-Rank Subspace Recovery

Face recognition has made great progress and becomes a hot spot by introducing the theory of low-rank subspace recovery. Based on the domestic and foreign relevant low-rank subspace recovery theory, this article proposes new methods of face recognition on the basis of low-rank subspace.Firstly, face image often has illumination, occlusion, or noise pollution problem. It affects the low-rank structure. In this article, we propose a multiple subspace sparse representation face recognition algorithm based on low-rank subspace recovery. First block the training sample, then reconstruct the training dictionary by using the block image.Then use low-rank decomposition to get low-rank matrix and sparse matrix error.After that, get low-rank matrix mapping matrix by principal component analysis, and map the training dictionary and testing dictionary to the mapping matrix. Finally, use maximum probability of partial ranking to get the classification results.Secondly, the sparse error which get from the low-rank subspace in recovery can effectively predict the test samples. So it proposes a joint discriminative dimensionality reduction and dictionary learning sparse representation face recognition algorithm based on low-rank subspace recovery. First of all, construct new training dictionary using low-rank decomposition get low rank matrix and sparse matrix error.Then using dimension reduction and dictionary learning learn a new training dictionary and mapping matrix. Finally, map the test sample and sparse coding in the new learning dictionary to get the result.Finally, in reality face image often has illumination and shadow problem, it makes the processing of the reconstruction error too large.This paper put forwards a face recognition algorithm based on the kernel feature space of low-rank subspace recovery.First of all, deal training sample with the low-rank decomposition in order to get low rank and sparse error part. Then use principal component analysis get mapping matrix.Map the training samples and testing samples in the mapping matrix. Finally, get the result in kernel feature space.