Constraint-based sparsity preserving projections and its application on face recognition

Aiming at the deficiency of supervise information in the process of sparse reconstruction in Sparsity Preserving Projections (SPP), a semi-supervised dimensionality reduction method named Constraint-based Sparsity Preserving Projections (CSPP) is proposed. CSPP attempts to make use of supervision information of must-link constraints and cannot-link constraints to adjust the sparse reconstructive matrix in the process of SPP. On one hand, CSPP obtains the high discriminative ability from supervised pairwise constraint information. On the other hand, CSPP has the strong robustness performance, which is inherited from the sparse representation of data. Experimental results on UMIST, YALE and AR face datasets show, in contrast to unsupervised SPP and existing semi-supervised dimensionality reduction method on sparse representation, our algorithm achieves increase in recognition accuracy based on the nearest neighbour classifier and promotes the performance of dimensionality reduction classification.