A Review of EEG-Based User Authentication: Trends and Future Research Directions

Abstract

The employment of Electroencephalography (EEG) in the User Authentication (UA) scientific research has recently unlocked state-of-the-art experimentation, aiming at identifying and authenticating individuals given their brainiac activity within specific contexts of use. Indeed, utilizing EEG signals that are derived from brainiac activities can be used for tackling existing UA security threats such as shoulder surfing, thus providing a novel solution to contemporary security problems in traditional knowledge-based user authentication. In this survey, we aim to complete previous literature surveys by providing a systematic classification and presentation of existing research that is based on the following pillars: a) the user experimental setup, with an emphasis on the applied EEG-acquisition protocols (e.g., rest, external stimuli driven, mental and hybrid); b) the artificial intelligence techniques employed and finally c) the security and privacy preservation aspects. The reviewed papers cover a broad area of experiment protocols and various algorithms used for EEG signal classification. Moreover, most cited works include results from more than one experiments with different approaches and configurations. This leads to a discussion on best practices for EEG-based User Authentication and conclusions suggesting future research directions that consists, among others, of considering homomorphically encrypted biometric templates for information leakage prevention through federated learning approaches in decentralized architectures. We anticipate that the present literature review will provide a roadmap for future research by considering efficiently and effective EEG-based User Authentication methods while at the same time preserving privacy.

Robust sub-network fingerprints of brief signals in the MEG functional connectome for single-patient classification

“…We instead took unique pair values in the 30 s snapshot of the functional connectome, resulting in a 45-component feature vector. Our deep learning model followed a basic architecture: dual-objective regression and binary classification (see Supplementary Tables 3,4,5). It was purposefully lightweight for reproducibility and simplicity.…”

“…Person re-identification from brain fingerprints has been studied extensively, particularly across Electroencephalography (EEG) [3] and functional-magnetic-resonance-imaging (fMRI) [4]. Studies in EEG finger-printing primarily focus on biometrics applications [5] via machine learning with signal processing methods. In contrast, advances in fMRI fingerprinting have focused on the functional connectome [4] and helped the understanding of several areas of neuroscience, such as cognition [6].…”

Distraction descriptor for brainprint authentication modelling using probability-based Incremental Fuzzy-Rough Nearest Neighbour

“…Furthermore, they are also applied in assistive, rehabilitative, and entertainment applications as the basis for Brain–Computer Interface (BCI) and Brain–Machine Interface (BMI). Despite widespread interest in clinical applications, the utilization of brain signals such as Electroencephalogram (EEG) is used as a biometric modality for person authentication or person identification [ 1 – 3 ]. EEG signal is an outstanding biometric modality with several benefits.…”

Cryptographic Algorithm Designed by Extracting Brainwave Patterns

“…When designing a system for authentication, it is necessary to account that this approach does not allow for password reset or recovery. On a large scale, there are two such protocols used for acquisition, namely Rest Eyes Open (REO) and Rest Eyes Closed (REC) [12,35]. The main disadvantages of this protocol are the sensitivity of external stimuli, which can disturb users' attention, or the modification of signals generated, and artifacts generated by controlled or uncontrolled movements, such as blinking, as well as significantly affecting the accuracy of the system [3,[36][37][38].…”