We show that the electroencephalogram (EEG) contains individual-specific information, which can be extracted and used for identity verification. We applied both linear (Auto Regressive) and nonlinear (Generalized Autoregressive Conditional Heteroskedasticity) models to 16 second frames of one channel EEG down-sampled to frequency 60 Hz. We used parameters of this model as the features and employed a Mahalanobis distance-based classifier for identity verification of individuals. The initial testing for 10 persons using measurements distributed over a long period of time reached 88% of correct verifications, showing the potential of our approach, as well as indicating that the results are time invariant.

Date Published: 12 October 2006
PDF: 9 pages
Proc. SPIE 6347, Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2006, 634722 (12 October 2006); doi: 10.1117/12.714567

Published in SPIE Proceedings Vol. 6347:
Photonics Applications in Astronomy, Communications, Industry, and High-Energy Physics Experiments 2006
Ryszard S. Romaniuk, Editor(s)