Algorithms for Basis Sparsity Non-negative Matrix Factorization

Abstract

In this paper, we proposed a method called basis sparsity non-negative matrix factorization, for decomposing multivariate matrix into sparse nonnegative components. The algorithm combining the original NMF formulation with sparseness constraint to make the basis matrix reaches to a more sparseness degree, and with local features clearer. By controlling the parameter, we can obtain the results of the basis matrix with different sparseness degree. Experiments have been done on the ORL database and the results show the algorithm has a good sparse ability.