Low rank analysis of eye image sequence– A novel basis for face liveness detection

Abstract

The security of the face recognition technology has attracted more and more attention because of the wide applications of this technology. A lot of studies on face liveness detection have been performed. In this paper, we cast the face liveness detection problem as a classification problem to distinguish the images of true faces and photo samples based on the rank analysis of sample matrices. We assume that the rank of the true face sample matrix is much higher than that of the photo sample matrix under an ideal situation. If we denoise the real world samples and convert them into pure samples, we can find a well boundary, that is, a basis for liveness detection. Experiments are conducted on the NUAA imposter database to verify the efficiency of the proposed method.