Study of continuous user authentication system based on EEG signals with the use of acoustic stimuli
Loading...
Date
2026-02-09
Authors
Κολόκας, Ηρακλής
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
EEG-based biometrics have gained increasing attention as a potential authentication method,due to their resistance to forgery and spoofing. However, many studies focus on performance, overlooking real-world applicability. This thesis studies an EEG-based authenticationsystem that utilises three pure-tone acoustic stimuli (500 Hz, 2000 Hz, 4000 Hz) to elicit stableand repeatable neural responses,with the goal of designing and evaluating an authentication system applicable to real-world use. The EEG data used in this study were collectedfrom 13 volunteer participants. They were preprocessed and segmented based on the onsetof the stimuli. 2 different feature extraction approaches were evaluated in this study: manual feature extraction on the time- and frequency-domain and embedding-based featureextraction using a lightweight convolutional neural network (EEGNet). Both approaches were evaluated under closed- and open-set scenarios. The closed-set scenario assumesusers are known and it is used to evaluate user separation, through multi-class classification,while the open-set scenario considers that both known and unknown users can try to access the system. The open-set relied on a template- and threshold-based system that comparedusers using cosine similarity and was evaluated by authentication specific metrics. Theresults of the research show encouraging evidence and provide the opportunity for furtherstudy and experimentation in this specific research area.
Description
Keywords
Electroencephalography (EEG), Acoustic stimuli, Authentication, Closed-set evaluation, Open-set evaluation, Feature extraction, Machine learning, Brain-computer interface