Face Recognition Based on Structural Incoherence and Low Rank Projection

Abstract

To solve the problem that both training and test samples are corrupted due to occlusion and disguise during face recognition, a new method which is based on low rank matrix recovery with structural incoherence and low rank projection (LRSI_LRP) is presented. First, the training images are decomposed into a set of clean images and sparse errors via LRSI, and the derived clean images from distinct classes are forced to be as independent as possible by introducing structural incoherence regularization term into robust PCA. Then a low-rank projection matrix is learned based on the original training images and the recovered clean ones, and this low-rank projection matrix can correct corrupted test samples by projecting them onto their corresponding underlying subspaces. Finally, the corrected test samples are classified based on sparse representation-based classification (SRC). Experimental results on AR and Extended Yale B databases verify the efficacy and robustness of the proposed method.