An alternative formulation of kernel LPP with application to image recognition

Locality preserving projections (LPP) is a new subspace feature extraction method which seeks to preserve the local structure and intrinsic geometry of the data space. As the LPP model is linear, it may fail to extract the nonlinear features. This paper proposes to address this problem using an alternative formulation, kernel locality preserving projections (KLPP). Our algorithm consists of two steps: kernel principal component analysis (KPCA) plus LPP. We provide an outline for implementing KLPP. Experiments on the ORL face database and PolyU palmprint database demonstrate the effectiveness of the proposed algorithm.