Towards an efficient and accurate EEG data analysis in EEG-based individual identification

Abstract

Individual identification plays an important role in privacy protection and information security. Especially, with the development of brain science, individual identification based on Electroencephalograph (EEG) may be applicable. The key to realize EEG-based identification is to find the signal features with unique individual characteristics in spite of numerous signal processing algorithms and techniques. In this paper, EEG signals of 10 subjects stay in calm were collected from Cz point with eyes closed. Then EEG signal features were extracted by spectrum estimation (linear analysis) and nonlinear dynamics methods and further classified by k-Nearest-Neighbor classifier to identify each subject. Classification successful rate has reached 97.29% with linear features, while it is only 44.14% with nonlinear dynamics features. The experiment result indicates that the linear features of EEG, such as center frequency, max power, power ratio, average peak-to-peak value and coefficients of AR model may have better performance than the nonlinear dynamics parameters of EEG in individual identification. © 2010 Springer-Verlag.