Global Sparse Representation Projections for Feature Extraction and Classification

Abstract

In this paper, we propose a novel supervised learning method called Global Sparse Representation Projections (GSRP) for linear dimensionality reduction. GSRP can be viewed as a combiner of sparse representation and manifold learning. But differing from the recent manifold learning methods such as Local Preserving Projections (LPP), GSRP introduces the global sparse representation information into the objective function. Since sparse representation can implicitly employ the "local" structure of the data by imposing the sparsity prior, we take advantages of this property to characterize the local structure. By combining the local interclass neighborhood relationship and sparse representation information, GSRP aims to preserve the sparse reconstructive relationship of the data and simultaneously maximize the interclass separability. Comprehensive comparison and extensive experiments show that GSRP achieves higher recognition rates than the state-of-the-art techniques such as LPP and Sparsity Preserving Projections (SPP).