Robust principal component analysis for recognition

Abstract

Recently, exactly recovering the intrinsic data structure from highly corrupted observations, which is known as robust principal component analysis (RPCA), has attracted great interest and found many applications in computer vision. Previous work has used RPCA to remove shadows and illuminations from face images. To go further, this paper introduces a method to use RPCA directly for recognition. And the inexact Augmented Lagrange Multiplier algorithm (ALM) is used to solve the RPCA problem. We actually utilize RPCA to reconstruct the testing sample from the training samples and compare the reconstructed one with the original one to do classification. Although the method is not very complicated, through experiments on some face databases we can see that it has better performance compared with some existing methods, especially under rigorous circumstances of occlusions and illuminations. © 2013 Springer-Verlag Berlin Heidelberg.