Video-based face recognition using the POEM descriptor

Numerous methods are proposed in describing and analyzing faces in videos using spatiotemporal operators and they have obtained very noteworthy results. This paper proposes three new operators, POEM-TOP, VPOEM and AMVs+LBP-TOP, for video-based face recognition. These operators use gradient orientation and gradient magnitude of video frames for video description and feature extraction. Robustness to uniform illumination variations and computational simplicity are among the benefits of the proposed operators. Experiments are applied on two standard databases, Honda/UCSD and VidTIMIT, which are provided for video-based face recognition. The experimental results show the effectiveness of our proposed methods.