City Research Online

Abstract

Security authentication is the process of confirming an individual’s identity. In terms of data security, authentication technology has played a crucial role for many years. However, there are still limitations with current standard biometric authentication technologies. The advancement of technology has resulted in the production of a variety of distinct devices that currently have the capacity to reproduce human biometrics and are, for the most part, both recognisable and touchable. Therefore, to overcome the limitations of existing biometric authentication systems, a new biometric method is necessary. Recently, various brain computer interface (BCI) applications have made use of human brain signals as a form of biometric identification. Electroencephalography (EEG) is a standard way to study how brain activity changes over time, especially when it comes to cognitive and disease-related activity. For the last two decades, researchers have studied EEG’s potential as a biometrics modality. However, its feasibility in the real world has not yet been determined, primarily due to issues related to collectability, reproducibility, and the permanence over time.

To overcome these issues and extend the effectiveness of EEG authentication, this thesis addresses the existing shortcomings of EEG biometrics and potential solutions organised in two main parts. Part 1 explores novel unsupervised learning approaches in noise removal by using variational autoencoder (VAE), extracts features efficiently with deep autoencoder and MiniRocket, then combines convolutional neural networks (CNN) and long-short term memory (LSTM) to create a hybrid model. Using publicly available EEG datasets, the proposed approaches are tested and shown to perform better than state-of-the-art methods for EEG-based authentication. Part 2 proposes an in-ear EEG-based biometrics system that can be deployed quickly, that is simple to use, can be done discreetly, and does not need the presence of specialised personnel or the use of bulky equipment which meets the collectability requirement. Moreover, in contrast to the vast majority of existing research, we take into account data collected from multiple subjects over the course of multiple recording days which solve the reproducibility issue. A dataset from 17 subjects over two different sessions was recorded (IEEMID); MiniRocket features are used to extract the features efficiently and then forwarded to linear classifier.

A thorough evaluation of multiple recordings and proof-of-concept studies in this thesis demonstrates the viability of the proposed frameworks. The wearable in-ear EEG paradigm resolves the critical issues with robustness associated with current use of EEG in authentication, and it promises reliable EEG applications in real-world settings.

Publication Type:	Thesis (Doctoral)
Subjects:	R Medicine > RC Internal medicine > RC0321 Neuroscience. Biological psychiatry. Neuropsychiatry T Technology > T Technology (General)
Departments:	School of Science & Technology > Engineering School of Science & Technology > School of Science & Technology Doctoral Theses Doctoral Theses

Export

Downloads