Information Technology Journal1812-56381812-5646Asian Network for Scientific Information10.3923/itj.2010.1719.1724YangHongli HeGuoping 8201098In this study, a novel algorithm is presented for dealing with the online data based on nonnegative matrix factorization model. Nonnegative Matrix Factorization (NMF) is a promising approach for face recognition in that it is capable of extracting the local features by factorizing the nonnegative matrix into two nonnegative matrices. However, there are two major weaknesses in almost all the existing NMF based methods. The first shortcoming is that the computational cost is high for large matrix factorization, it also needs more memory to save the huge data. The other is that it must conduct repetitive learning when the data samples are updated. To overcome these two limitations, a novel Online Nonnegative Matrix Factorization (ONMF) algorithm for online face recognition is presented in this study. The ONMF algorithm can not only deal with the incremental nonnegative matrix factorization, but also can deal with the decremental nonnegative matrix factorization which has never been considered in other study. Two face databases, namely ORL and FERET face database, are selected for evaluation. Compared with the conventional standard NMF, the method in this study gives the better performance both in computational costs and other aspects.]]>Lee, D.D. and H.S. Seung,1999Lee, D.D. and H.S. Seung,2000Wen-Sheng, C., B. Pan, B. Fang, M. Li and J. Tang,2008Guillamet, D. and J. Vitria,2002Guillamet, D. and J. Vitria,2003Bucak, S.S. and B. Gunsel,2009Rebhan, S., W. Sharif and J. Eggert,2008Cao, B., D. Shen, J.T. Sun, X. Wang, Q. Yang and Z. Chen,2007Pan, B.B., W.S. Chen and C. Xu,2008Li, G., S. Zhang, W. Wang and B. Shi,2010Bucak, S.S. and B. Gunsel,2007Hoyer, P.O.,2004