Incremental nonnegative matrix factorization with sparseness constraint for image representation

Abstract

Nonnegative matrix factorization (NMF) is a powerful method of data dimension reduction and has been widely used in face recognition. However, existing NMF algorithms have two main drawbacks. One is that the speed is too slow for large matrix factorization. The other is that it must conduct repetitive learning when the training samples or classes are incremental. In order to overcome these two limitations and improve the sparseness of the data after factorization, this paper presents a novel algorithm, which is called incremental nonnegative matrix factorization with sparseness constraint. By using the results of previous factorization involved in iterative computation with sparseness constraint, the cost of computation is reduced and the sparseness of data after factorization is greatly improved. Compared with NMF and INMF, the experimental results on some face databases have shown that the proposed method achieves superior results.