Locality Sensitive Sparse Concept Coding and Its Application to Image Representation

Matrix Factorization is very effective image representation approach in pattern recognition. The traditional matrix factorization methods cannot capture the intrinsic structure information.In this paper, a novel method, called Locality Sensitive Spare Concept Coding (LSSCC) , is proposed which can capture the intrinsic geometrical structure and discriminate information in basis learning. Therefore, it can find a basis set capturing high-level semantics information of the data.And the coefficients are obtained when the samples are sparse representation on the basis vectors.Finally, the samples are represented and classified.The clustering experiments on the COIL 20 and ORL database demonstrate the proposed algorithm can effectively improve the accuracy and normalized mutual information in clustering and verify the effectiveness compared to other matrix factorization algorithm.