Improved NMF algorithm based color face recognition

Class information is very important to face recognition. A novel class-dependent color face recognition approach is proposed in this work. Color face is first decomposed into three different components according to three color channels. Non-negative matrix factorization algorithm is improved by imposing one more constraint on factorization matrices. And class information of face samples is integrated into the algorithm for feature extraction on each color channel respectively. 3D active deformable model is exploited to generate realistic virtue color face images to improve the quality of feature extraction. Experimental results on CVL and CMU PIE face databases verify effectiveness of the proposed approach.