In this paper, we propose a Non-negative Sparsity Preserving Projections (NSPP) algorithm and apply the proposed algorithm to face recognition. We propose a more reasonable method of constructing the weight matrix and the coefficients of the weight matrix are all non-negative. This method is more consistent with the biological modeling of visual data and often produces much better results for data representation. Experimental results have shown that NSPP algorithm outperforms Locality Preserving Projections and Sparsity Preserving Projections on both ORL and FERET face database. The weight matrix is non-negative and posses more sparsity, which can enhance recognition performance in the projected low-dimensional subspace.
展开▼
