EEG biometric identification: Repeatability and influence of movement-related EEG

This paper describes use of EEG signal as biometric characteristic for person identification. We focus on the problem of repeatability of the identification process, and influence of the movement-related EEG on results of identification. Used database of EEG signals consists of two sessions, obtained approximately one year apart. We use Frequency-Zooming Auto-Regression modeling and Mahalanobis distance-based classifier for classification of EEG segments, which leads to subject identification with success rate for single session identification up to 98%. When the earlier session is used for classifier training and the later session for testing, the highest success rate with our identification algorithm is 87.1%. Experiments show that use of the movement-related EEG leads to better identification results.