Exploiting Mallows Distance to Quantify EEG Distribution for Personal Identification

Electroencephalogram (EEG) activity from the brain is a promising biological marker that can serve as personal identification. Despite substantial efforts, it still remains unsolved problems to quantify EEG feature distribution for brain biometrics due to the complexity of the brain. In this study, we attempt to tackle EEG-based identification challenges by exploiting a novel distribution model. The distribution dissimilarity is measured by Mallows distance, a cluster similarity sensitive distance that is robust to signal noises. Specifically, EEG signals are decomposed through several statistical feature extraction methods, autoregressive (AR) model, discrete wavelet transform (DWT), and fast Fourier transform (FFT). With the dataset obtained from the real-world application, our proposed system achieves the f-score accuracy of 96.18% and half total error rate of 2.223%, which demonstrates the feasibility and effectiveness of utilizing EEG biometrics for personal identification applications.