Undergraduate Final Thesis Presentation. Guided by Dr. Lingling Yang, SYSU Computational Medical Imaging Lab
2017.07-2018.02
Electroencephalography (EEG) has been widely investigated for person authentication because of its advantages of being difficult to fake, impossible to observe or intercept, and unique. In this study, we proposed a new person authentication system based on EEG signals. The self-paced reaching task was applied as it was a natural and common human daily task. Power spectral density (PSD) of delta, theta, alpha and beta bands were extracted from EEG signals of channel C3 as features. After comparing different classification algorithms’ effect, a support vector machine was applied as a classifier. Our study showed the superior performance of alpha and beta band PSD features, compared to delta and theta bands. The time course characteristics of alpha and beta bands were also studied. It was showed that alpha band PSD features in [-2 1] s and the beta band PSD features in [-1 0] s and [0 1] s achieved the best results. Moreover, the AUC of the person authentication performance decreased when the combined features of PSD and AR parameters were applied. Overall, the best average accuracy of 78.0% was achieved, which demonstrates the possibility of using a reaching task for person authentication.