Investigate the potential of EEG signals for biometric authentication: A power spectral density approach

In the modern world, authentication and access control mechanisms are required for many activities. Traditional methods of identification include token-based systems, such as passport or driver license or knowledge-based systems, such as passwords or PIN-codes. A more advanced approach for authentication is Biometrics. Biometrics is the science of measuring and analyzing certain unique human body characteristics for authentication purposes. These unique human body characteristics are called Biometric Identifiers. Some common examples of biometric identifiers include fingerprints, DNA, face recognition, iris recognition, gait, typing rhythm, etc. The advantage of using Biometrics for authentication is the use of some part of the body for authentication, so the individual does not need to carry objects like identification cards or remember passwords. Also, Biometrics is considered more fraud resistant than conventional techniques.;An emerging approach in biometrics is the EEG-based Cognitive Biometrics, where the brain's electric response to some stimuli is utilized. EEG results from the electrical activity due to the ionic current flows within the neurons of a functioning brain. Brain activity of each person is unique and EEG signals can be used as a potential biometric identifier for authentication purposes. An EEG-based biometric system might be more fraud resistant than conventional biometric systems, since the brain activity is secure and cannot be replicated forcefully.;The aim of this work is to investigate the potential of using EEG signals as a biometric identifier for biometric authentication and also to investigate if power spectral density can be used as a unique feature of EEG signals for biometric authentication. EEG signals were recorded from eight healthy subjects while exposing them to six different visual stimuli to evoke emotional responses. Recorded EEG signals were processed and analyzed using signal preprocessing techniques, such as power spectral density (PSD) estimation, statistical analysis of differences (Kruskal-Wallis test), and classification of subjects using Euclidean distance and Artificial Network classifiers. EEG data were analyzed separately for all seven EEG rhythms to determine which rhythms can be used for better classification. With the accuracy up to 96.42%, Kruskal-Wallis test results confirmed that PSD estimates are different for different subjects and can be used as a unique EEG feature for Biometric authentication. The highest classification accuracy of 87.5% was achieved for ?1 EEG rhythm (8-10 Hz) while using the Artificial Neural Network classifier and ?2 EEG rhythm (10-14 Hz) while using the Euclidean distance classifier. The latter indicates that the proposed approach allowed successful classification of 7 out of the 8 subjects using the averaged PSD of their ?1 and ?2 rhythms EEG.