Abstract: Non-negative matrix factorization (NMF) is a new matrix decomposition method based on the part of the study,which has a reflection of human thinking partial constitute the overall concept. It only find two nonnegative matrices whose product can approximate the nonncgative data matrix without considering the geometric structure and the discriminative information in the data. We presented a local sensitive nonnegative matrix factorization for dimensionality to overcome the disadvantage, which preserves not only the nonnegativity but also the geometric structure and discriminative information of the data. An efficient multiplicative updating procedure was produced, and its convergence was gua-ranteed theoretically. Experiments on ORL and Yale face recognition databases demonstrate that proposed method outperforms many existing dimensionality reduction methods.

Key words: Nonnegative matrix factorization,Local sensitive analysis,Discriminative information,Ueometric structure