关键词: 子空间降维,稀疏约束,非负矩阵分解,增量学习

Abstract: Non-negative matrix factorization is a useful method of subspace dimensionality reduction．However,with the increasing of training samples,the computing scale of non-negative matrix factorization increases rapidly．To solve this problem and improve the sparseness of the data obtained after factorization as well,an incremental learning algorithm of non-negative matrix factorization with sparseness constraints was proposed in this paper．Using the results of previous factorization involved in iterative computation with sparseness constraints,the cost of the computation is reduced and the sparseness of data after factorization is highly improved．Experimental results on both ORL and CBCL face databa-ses show that the proposed method is effective on dimensionality reduction.

Key words: Subspace dimensionality reduction,Sparseness constraints,Non-negative matrix factorization,Incremental learning