Bregman iteration algorithm for sparse nonnegative matrix factorizations via alternating l 1-norm minimization

Sparse nonnegative matrix factorizations can be considered as dimension reduction methods that can control the degree of sparseness of basis matrix or coefficient matrix under non-negativity constraints. In this paper, by exploring the sparsity of the basis matrix and the coefficient matrix under certain domains, we propose an alternative iteration approach with l (1)-norm minimization for face recognition. Moreover, a modified version of linearized Bregman iteration is developed to efficiently solve the proposed minimization problem. Experimental results show that new algorithm is promising in terms of detection accuracy, computational efficiency.