Aiming at the problem that non-negative matrix factorization（ NMF） has certain sparsity

but it is not enough for effective classification

in order to obtain high-dimensional data structure information and hidden information missing in the feature extraction process

improve the low rank and sparsity of the non-negative matrix factorization

a non-negative matrix factorization model based on latent low-rank representation（ NLatMF） is proposed. Non-negative low-rank part and latent parts of the original data extracted by latent low-rank algorithm

applied to NMF

which solves classification problem more effectively. The improved model is used for image classification and the results show that the new model can improve the recognition rate effectively by simulating on the face database such as Yalefaces.