Two-Dimensional Weighted and Locality Preserved Discriminant Analysis for Face Recognition

Abstract

In this paper, a novel method for face recognition, called two-dimensional weighted and locality preserved discriminant analysis (2D-WLPDA) is proposed. The new algorithm is developed based on three techniques: (1) locality preserved embedding, by embedding nearest-neighbor graphs which characterize the within-class compactness of the same class samples, 2D-WLPDA discovers the submanifold of images space; (2) image based projection which can avoids the small sample problem and improves the computation efficiency;(3) weighting contributions of individual class pairs which alleviates the overlap of neighboring classes in Fisher criterion for a k-class problem with k>2. We experimentally compare 2D-WLPDA to other feature extraction methods, such as 2D-LDA, 2D-PCA and 2D-DLPP, 2D-WLPDA has better recognition performance.