Face Recognition Based on Discriminant Sparsity Preserving Embedding
Xiao Ma · Acta Automatica Sinica · 2014
Motivated by the recent rapid development of sparse representation(SR) on feature extraction and dimensionality reduction of high-dimensional data such as face images, an improved version of sparsity preserving projection(SPP), named discriminant sparsity preserving embedding(DSPE), is proposed in this paper. Through solving a least square problem, the sparse weight of SPP is updated and the discriminant sparse weight which actually reflects the discriminant information is obtained. Then the low-dimensional feature subspace of the original high-dimensional data is evaluated by best preserving such sparse weight relationship. DSPE is a linear supervised learning method which can deal with high-dimensional data efectively by introducing discriminant information. The efectiveness of the proposed method is verified on four popular face databases(ORL, Yale, Extended Yale B, and CMU PIE) with promising results.