The brainwave signal has recently been investigated for person identification and verification due to its advantages over traditional means. Most research on brain signals focuses on BCI or clinical applications. However little work has been done in brainwave-based person recognition or emotion recognition, and none has been done in brainwave-based human characteristics classification. Although numerous pre-processing, feature extraction and classification methods have been proposed and explored for BCI systems, none of them has been accepted as the best method. This research project focuses on using Electroencephalography (EEG) as a new biometric to build person recognition systems. These systems include EEG-based person recognition that can identify or verify a person using a given unknown EEG signal, and EEG-based human characteristics classification that can classify a person in to an age group, gender group, or emotion group based on his/her EEG signal. This research also proposes a new feature extraction method based on speech analysis, and a new modelling method based on support vector machines to enhance the performance of those EEG-based recognition systems. The following feature extraction methods from the speech domain are proposed: speech recognition features, speaker recognition features, and speaker characteristics features. Models proposed for human characteristics classification include D-SVDD, TD-SVDD,MTD-SVDD,R-SVDDs and MR-SVDDs. Models proposed for person verification are MS-SVDD UBM,FMS-SVDD UBM,M3S-SVDD UBM and FM3SSVDD UBM. Evaluation experiments performed on the Australian EEG,EEGMMIDB, Alcoholism, Graz and DEAP datasets show better results for most of the proposed techniques than do traditional techniques.
On EEG-based person recognition and human characteristics classification
Nguyen, P. (Author). 2015
Student thesis: Doctoral Thesis