Multiview Dimension Reduction Based On Sparsity Preserving Projections

In this paper, we focus on boosting the subspace learning by exploring the complimentary and compatible information from multiview features. A novel multi-view dimension reduction method is proposed named Multiview Sparsity Preserving Projection (MSPP) for this task. MSPP aims to seek a set of linear transforms to project multiview features into subspace where the sparse reconstructive weights of multi-view features are preserved as much as possible. And the Hilbert Schmidt Independence Criterion (HSIC) is utilized as a dependence term to explore the compatible and complementary information from multiview features. An efficient alternative iterating optimization is presented to obtain the optimal solution of MSPP. Experiments on image datasets and multi-view textual datasets well demonstrate the excellent performance of MSPP.