Robust feature modeling for face authentication in smart device

Ao Li, Xin Liu, Hailong Jiang

Abstract: As smart devices become widespread, security has emerged as a crucial problem for users. Biometrics is an effective means of user authentication that has been applied to smart devices. Face authentication has been widely studied in this context because of its short interaction time and quick authentication. In this paper, we propose robust feature modeling for face authentication in smart devices. Our approach uses coefficients of low-rank representation as a measure of similarity between samples. Label regression and reconstructed label information are then used to supervise feature subspace learning. We added three types of noise and interference to three public face datasets to simulate potential problems in face authentication on smart devices. The results of the experiment show that our proposed approach can achieve better authentication performance than conventional methods.