Graph Embedding Algorithms Based on Neighborhood Discriminant Embedding for Face Recognition

Abstract

This paper explores the use of a series of Graph Embedding (GE) algorithms based on Neighborhood Discriminant Embedding (NDE) as a means to improve the performance and robustness of face recognition. NDE combines GE framework and Fishers criterion and it takes into account the Individual Discriminative Factor (IDF) which is proposed to describe the microscopic discriminative property of each sample. The tensor and bilinear extending algorithms of NDE are proposed for directly utilizing the original two-dimensional image data to enhance the efficiency. The common purpose of our algorithms are to gather the within-class samples closer and separate the between-class samples further in the projected feature subspace after the dimensionality reduction. Furthermore, another informative feature extraction method called Circular Pixel Distribution (CPD) is applied to enhance the robustness of the 2-D algorithm. Experiments with the Olivetti Research Laboratory (ORL) face dataset are conducted to evaluate our methods in terms of classification accuracy, efficiency and robustness.