A Manifold Learning Algorithm for Video-based Face Recognition

摘要

Our research focus is to fully utilize both spatial and temporal information in video to overcome the difficulties in the video-based face recognition, such as large variations of face scale, radical changes of illumination, low resolution of face images in video and pose, as well as occasionally occlusion of different parts of faces. This paper proposed a novel video-based face recognition algorithm using clustering (CVLPP), which can discover more special semantic information hidden in video face sequence, simultaneously uncover the intrinsic manifold structure information to preserve discriminative inherent features. We also compare our approach with other algorithms on our own Video Database. The experimental results show that CVLPP can get higher recognition accuracy rate for video-based face recognition.