Improved non-negative matrix factorization and its application to face recognition
Ruixia Yang · Guangdian gongcheng · 2007
In order to improve the robustness to external factors such as illumination and pose,a novel Non-negative Matrix Factorization(NMF) algorithm using discriminant information is presented. The proposal employs linear discriminant analysis in the features derived from NMF. In this way,a subspace having the maximum intersubject variation and the minimum intrasubject variation is established. The feature is discriminated in subspace. The proposed method is applied to face recognition using the well-known Havard database and Umist database. Experiment results indicate that this method is robust to large illumination and pose variations,it greatly enhances the performance of NMF,and the recognition rate increases more than 20%.