An EEG-based biometric system

Abstract

Electroencephalography (EEG) is the non-invasive way of recording the electrical activity in the brain by placing electrodes on different parts of the human scalp. EEG signals are used in two main areas: medical area and brain computer interfacing (BCI). EEG is used as a diagnostic tool for identifying medical abnormalities such as epilepsy and brain tumours. Recent studies contributed in the use of EEG as a biometric trait. The EEG is unique making it a good candidate to use as a biometric trait. Moreover, EEG-based biometric system offers several benefits over other biometric systems. Since the EEG is a summation of the electrical potentials caused by the neurons, it is very difficult for an attacker to acquire the EEG data and feed it to the system. This will result in a more robust system against sensor spoofing offering a higher level of security. This system can be also used by people with severe disabilities such as missing hands and fingertips. The EEG-based biometric system proposed in this project uses a new experimental protocol that elicits a type of response in the brain that has never been analysed before in terms of biometry. This response is known as the steady state visually evoked potential (SSVEP). A literature review was carried out to investigate the techniques used for biometry in terms of feature extraction and classification algorithms. Comparing the methods used in the different studies leads to a selection of the appropriate methods to implement in this project. Another important aspect that emerged from the literature review is the importance of the ageing effect in biometry and this was tackled by considering data from two sessions with a one-week time span. Since SSVEPs in biometry has never been used so far, the appropriate frequencies and duty cycle had to be analysed within each individual before discriminating between the subjects. After selecting the best frequencies and duty cycle which performed best across all the six subjects, the discrimination of the six subjects follows using different types of feature vectors. This was done to study which elements have the most relevant information to discriminate between the subjects. The best classification accuracy of 65 per cent was obtained when using the 25 per cent duty cycle and the stimulus frequency combination of 10Hz and 17Hz. This result can be further improved by implementing other feature extraction methods and classification algorithms.