An EEG-based biometric system

Abstract

Biometric systems have gained increased popularity in modern society since they provide an extra sense of security. A biometric system refers to a system that is capable of identifying an individual from a number of individuals by using specific biometric features. Standard biometric features used in common biometric systems include fingerprints, voice, and facial features. However, in recent years, studies considering electroencephalography (EEG) as a biometric feature have become more popular. The main advantage of using EEG in biometrics when compared to common biometric features is that it is not prone to spoofing due to the difficulty in replicating the signal, making the system much more secure. From previous studies, it was noted that very few considered the use of steady state visually evoked potentials (SSVEP) as biometric features, and no reasearch up to date considered the phase information from SSVEP. glsssvep are oscillatory responses in the EEG elicited when an individual is subject to a visual stimulus. On this basis, a study was conducted to increase the performance in biometric systems using magnitude information as biometric trait. Moreover, an initial investigation on the phase information in SSVEP signals was conducted to check viability of using phase information as biometric features. Data was collected from ten subjects, using three different stimulus frequencies. To investigate the effect of ageing in the EEG, data from each individual was recorded in three different sessions. From the initial study it was concluded that phase information carries distinctive properties, and remains significantly unchanged across time, which are very important characteristics in a biometric trait. Data recorded from across the three different sessions was used to test the biometric systems. The best overall classification accuracy for the biometric system using magnitude information was 53.8%. The best classification accuracy for one subject for the biometric system using phase information was 57%, when using only one feature for discrimination. The results do not show high accuracy results in identification of individuals. However these results can be improved further with better classification algorithms and larger feature vectors.