Manifold-based Face Gender Recognition for Video

Automatic face gender recognition for video has become one of the hottest topics in pattern recognition and machine learning nowadays. How to better utilize the space-time information hidden in video sequences to overcome the difficulties is the key point for video-based face gender recognition. In this paper, we propose a novel manifold-based algorithm called video-based face gender using tensor subspace analysis(VG-TSA), which can not only discover more special semantic information existed in video face, but also make full use of the intrinsic nonlinear structure information to extract discriminative lower-dimensional manifold features. Finally, we compare the proposed VG-TSA with other static algorithms on UCSD/Honda and our own video databases and achieve a better performance in video-based face gender recognition.