Quality filtering of EEG signals for enhanced biometric recognition

In this paper we present a biometric person recognition system based on EEG signals incorporating a novel strategy to find and utilize the most informative data segments using the concept of Sample Entropy. The users are presented with a stimulus that prompts a motor-imagery response. This is then measured using an array of EEG sensors. A sliding-window segmentation scheme and Wavelet Packet Decomposition are adopted for primary feature extraction before the quality measurement stage. The quality-filtered feature windows are then used to extract secondary features that are in turn classified using a linear discriminant classifier. The proposed system is tested using a publicly available EEG database and it shows that entropy filtering results in a significant improvement on performance. An average identification accuracy rate of more than 90% is achieved for 109 subjects using only eight electrodes, utilizing only the highest quality for each subject.