Conditional entropy approach to multichannel EEG-based person identification

Abstract

Person identification using electroencephalogram (EEG) as biometric has been widely used since it is capable of achieving high identification rate. Since single-channel EEG signal does not provide sufficient information for person identification, multi-channel EEG signals are used to record brain activities distributed over the entire scalp. However extracting brain features from multi-channel EEG signals is still a challenge. In this paper, we propose to use Conditional Entropy (CEN) as a feature extraction method for multi-channel EEG-based person identification. The use of entropy-based method is based on the fact that EEG signal is complex, non-linear, and random in nature. CEN is capable of quantifying how much uncertainty an EEG channel has if the outcome of another EEG channel is known. The mechanism of CEN in correlating pairs of channels would be a solution for feature extraction from multi-channel EEG signals. Our experimental results on EEG signals from 80 persons have shown that CEN provides higher identification rate, yet less number of features than the baseline Autoregressive modelling method.