New sparse subspace learning approaches for feature abstraction and recognition

Abstract

In this paper, we propose two novel sparse representation based dimension reduction approaches for feature abstraction and recognition: sparse local preserving projection (SLPP) and structural sparse local preserving projection (SSLPP). They are efficient in detecting the nonlinear features of the intrinsic manifold structure, also improving the interpretability of the projection. In addition, SSLPP promotes a more organized structural sparse pattern, overcoming the problem that just decreasing the cardinality may not be enough in some situations. Experiments in data classification and face recognition are carried out to verify the validity and effectiveness of the proposed methods.