Kernel LDP Based Discriminant Analysis for Face Recognition

Abstract

Locally discriminating projection (LDP) is a new subspace feature extraction method which takes special consideration of both the local information and the class information. As the LDP model is linear, it may fail to extract the nonlinear features. This paper proposes to address this problem using an alternative formulation, kernel locally preserving projection (KLDP). The proposed method consists of two steps: kernel principal component analysis (KPCA) plus LDP. An outline for implementing KLDP is provided. Experiments on the AR face database and Yale face database demonstrate the effectiveness of the proposed method.