EEG human biometric authentication using eye blink artefacts

This study proposes a new electroencephalography (EEG) biometric authentication for

humans based on eye blinking signals extracted from brainwaves. The brainwave signal has

been investigated for person authentication over the years because of its difficulties in

spoofing. Due to advancing low-cost EEG hardware equipment, it has recently been

significantly explored. Most studies in brainwave authentication focus on the use of

imagination and mental task to authenticate a subject. Such conventional approaches are

prone to the effect of human emotions and exercising, since this effect alters the brainwave

signal significantly, making such approaches to be less practical in the real world. This study

overcomes this limitation by introducing a new approach, where the effect of eye blinks on

the brainwave is used for authentication. The eye blink effect on the brainwave signal is

considered an artefact in EEG authentication and is usually removed at the pre-processing

stage. However, it holds properties that are ideal for use in authentication, and it is not prone

to human emotions and exercising, thus improving the practicality of brainwave

authentication. Brainwaves were recorded using Neurosky Mindwave Mobile 2 headset. The

NeuroSky blink detection algorithm was used to extract eye blinks and their properties from

the brainwaves. A new authentication algorithm is developed based on three (3) properties:

blink strength, blink time, and the number of blinks at a given time. The proposed

authentication algorithm matches the eye blinking properties stored in a database at the

enrolment stage against the one recorded at the authentication stage. The overall algorithm

results were calculated on a range of 0 – 100. A threshold value of 70 was used to authenticate

a subject. Three (3) experiments were conducted. In the first experiment, we evaluated the

performance of the proposed algorithm. The second experiment evaluated the effect of

emotions (Excitement, Calmness and Stress) on the proposed algorithm. The third experiment

evaluated the effect of exercising on the proposed algorithm. The performance of the

algorithm is measured using False Rejection Rate (FRR), False Acceptance Rate (FAR), and

Accuracy (ACC). Results showed a FAR value of 5% and an FRR value of 1%. The proposed

algorithm achieved an accuracy of 97%. These results show good performance. Results also

indicate that more complex patterns have low FAR and high FRR, while less complicated

xv

patterns have high FAR and low FRR. Results also show that human emotions and exercising

have no significant impact on the proposed approach.