Linear Discriminative Sparsity Preserving Projections for Dimensionality Reduction
Jianbo Zhang, Jinkuan Wang · 2018
Recently, some works have demonstrated that combining manifold learning or sparse subspace learning methods with global discriminative learning methods is an effective way to improve the performance for image recognition tasks. In this paper, to further exploit the robustness of those combinations, we propose a new algorithm called linear discriminative sparsity preserving projections (LDSPP). Different from the previous works which only consider within-class or between-class discriminative information, LDSPP takes both of them into account and adds them to sparse subspace learning model to establish objective function for dimensionality reduction. Experiments on Yale and ORL face image datasets demonstrate the effectiveness of the proposed method.