Study on characteristic dimension and sparse factor in Non-negative Matrix Factorization algorithm

Non-negative Matrix Factorization (NMF) algorithm and its variations have been successfully applied to many fields, but how to set the characteristic dimension value and the sparse factors value to improve recognition accuracy has been puzzling the researchers. Until now, it is regretful that the rigorous algorithm doesn´t appear. The purpose of this paper is not to improve existing NMF algorithm to improve recognition accuracy, but emphasis on investigating how sparse factors and the characteristic dimension affect recognition accuracy, and study how to set the optimization values to characteristic dimension and sparse factors respectively to obtain the optimization recognition accuracy. A platform for face recognition is built, and some experiments are carried out with the help of the platform, finally some directional conclusions are gained.