Enhanced Adaptive Locality Preserving Projections for Face Recognition

Abstract

In this paper, we address the graph-based manifold learning method for face recognition. The proposed method is called enhanced adaptive Locality Preserving Projections. The EALPP integrates four properties: (i) introduction of data label information and parameterless computation of affinity matrix, (ii) QR-decomposition for acceleration of the eigenvector computation, (iii) matrix exponential for solving the problem of singular matrix and (iv) processing of uncorrelated vector of projection matrix. EALPP has been integrated two techniques: Maximum Margin Criterion (MMC) and Locality Preserving Projections (LPP). Face recognition test on four public face databases (ORL, Yale, AR and UMIST) and experimental results demonstrate the effectiveness of EALPP.