Kernel Uncorrelated Local Fisher Discriminant Analysis and Its Application to Face Recognition

Abstract

Local Fisher Discriminant Analysis (LFDA) achieves high performance for face recognition. However, LFDA is still a linear technique and usually deteriorates because the basis vectors of LFDA are statistically correlated. In this paper, we propose a Kernel Uncorrelated Local Fisher Discriminant Analysis (KULFDA), which can exploit the nonlinear and statistically uncorrelated features. A major advantage of the proposed method is that every column of the kernel matrix is regarded as a corresponding sample. Then nonlinear features can be extracted by performing ULFDA the in kernel matrix. Experimental results on ORL and YALE databases demonstrate the effectiveness of the proposed algorithm.