Supervised local sparsity preserving projection for face feature extraction

Xiao-Yuan Jing,Sheng Li,Songhao Zhu,Qian Liu,Jingyu Yang,Jiasen Lu

DOI: https://doi.org/10.1109/ACPR.2011.6166662

2011-01-01

Abstract:In the sparse representation of a target sample, most nonzero coefficients belong to the neighbors of the target sample. Combining this observation with the theory of manifold learning, we propose a novel unsupervised feature extraction approach named local sparsity preserving projection (LSPP). LSPP sparsely reconstructs a target training sample from merely its neighbors, and seeks a subspace where the local sparse reconstructive relations among all training samples are preserved. To improve the discriminating power of LSPP, we further propose a supervised LSPP (SLSPP), which incorporates the class information of neighbor samples into local sparse representation. Experimental results on the AR and CAS-PEAL face databases demonstrate the effectiveness of LSPP and SLSPP, as compared with related feature extraction methods.