Spatially smooth subspace face recognition using LOG and DOG penalties

Abstract

Subspace face recognition methods have been widely investigated in the last few decades. Since the pixels of an image are spatially correlated and facial images are generally considered to be spatially smoothing, several spatially smooth subspace methods have been proposed for face recognition. In this paper, we first survey the progress and problems in current spatially smooth subspace face recognition methods. Using the penalized subspace learning framework, we then proposed two novel penalty functions, Laplacian of Gaussian (LOG) and Derivative of Gaussian (DOG), for subspace face recognition. LOG and DOG penalties introduce a scale parameter, and thus are more flexible in controlling the degree of smoothness. Experimental results indicate that the proposed methods are effective for face recognition, and achieve higher recognition accuracy than the original subspace methods. © 2009 Springer Berlin Heidelberg.