Supervised sparsity preserving projections for face recognition

Abstract

Sparsity preserving projection (SPP) is a recently proposed unsupervised linear dimensionality reduction method for face recognition, which is based on the recently-emerged sparse representation theory. It aims to find a low-dimensional subspace to best preserve the global sparse reconstructive relationship of the original data. In this paper, we propose a supervised variation on SPP called supervised sparsity preserving projection (SSPP). The SSPP method explicitly takes into account the within-class weight as well as between-class weight and assigns different weights to them, which attempts to strengthen the discriminating power and generalization ability of embedded data representation. The effectiveness of the proposed SSPP method is verified on two standard face databases (Yale, AR).