Exponential Neighborhood Preserving Discriminant Embedding for Face Recognition

As a manifold reduced dimensionality technique, neighborhood preserving discriminant embedding (NPDE) was proposed recently. But in most cases, NPDE has the so-called small-sample-size (SSS) problem. To address this problem, an exponential neighborhood preserving discriminant embedding (ENPDE) method is proposed in this paper. The main idea of ENPDE is that the matrix exponential is introduced to NPDE. ENPDE has two superiorities. Firstly, ENPDE avoids the SSS problem. Secondly, ENPDE has an effect to enlarge the distance between samples belonging to different classes in the neighborhood, and then the discrimination property is emphasized. The experiments are made on CMU-PIE and AR face databases, and ENPDE is compared with the PCA, LDA, EDA and NPDE methods. The experiment results show that, ENPDE is an efficient method and shows advantageous performance over the above methods.