Abstract:

A smooth non-negative matrix factorization with orthogonal exponent constraints is proposed. In this paper, the non-negative matrix is decomposed into the product of base matrix, column normalized smooth matrix and the coefficient matrix. At the same time, the objective function is added to the orthogonal exponent constraints, which ensures that the low dimensional characteristics are of the non-negativity and localization, reduces the decomposition error, and improves the ability of sparseness adjustment. The proposed method is used to data dimension reduction, feature sparseness comparison, occluded face recognition and video motion feature extraction. The experimental results show that the proposed method has a better performance than other similar methods.