Modified LNMF based color face recognition

Dimensionality reduction technique is very important to appearance-based face recognition algorithm. Compared with monochromatic face image, color face image which is composed of different color channels can provide more cues for recognition task. In this paper, a novel appearance-based recognition approach, modified local NMF based color face recognition, is proposed. Block diagonal matrix mode is employed to encode color information of different channels directly. Block diagonal constraint is plus into that of traditional local NMF to build modified local NMF for dimensionality reduction and feature extraction. 3D active deformable model is exploited to generate virtual face images for algorithm learning. Projective coefficients are adopted as facial features for recognition.