EEG based biometric authentication using new spectral features

From the viewpoint of user management, continuous authentication is effective. However, general biometrics such as fingerprint, iris, vein, and so on are not suitable for the continuous verification since they require conscious presentation of biometric data. As unconscious biometrics, to use a brain wave (electroencephalogram: EEG) has been proposed but conventional approaches require heavy computational load for feature extraction and verification and it becomes a problem for practical applications. In this paper, we propose new features: the concavity and convexity of spectral distribution in the alpha band and propose to fuse them with a spectral variance in verification. The proposed methods are achieved with low computational load. In our experiments using 23 subjects, EER of 11% is obtained and it is comparable with the performance of conventional approaches.