Multi-frame Approaches To Improve Face Recognition

Abstract

Face recognition from video sequences is becoming an important area of biometrics research. In this work, we explore different strategies to improve face recognition performance from video. We develop a good strategy to select the smallest number of frames to achieve a high level of performance. We apply Principal Component Analysis to identify suitable frames to represent the subjects. We demonstrate our approaches on our dataset, which uses three different cameras and is larger than any known research database of video face sequences. Finally, we compare our approach to an existing approach from UCSD [8, 9] and show that it performs slightly better than that approach (99% rank one recognition rate).