Automatic person authentication using fewer channel EEG motor imagery

Abstract

In today's world, there are different aspects of security in which appropriate computing technologies play an essential role. One such aspect is person's identification. While there are numerous ways to identify a person, from using finger prints to using face recognition; most of them exhibit, on one way or the other, unacceptable levels of reliability. On the other hand, recent developments in brain computer interfaces (BCI), using Electroencephalogram (EEG) signals have been emerging as a feasible option for identification systems. Current EEG based authentication systems use more than 8 up to even 60 electrodes placed on the scalp to record data. In this work, we propose and analyze an approach in which person's identification is achieved by measuring the EEG signals that the person generates while imagining simple motor movements, and which requires as few as 2 to 6 channel electrodes. The system uses the Short Time Fourier Transform (STFT) for extraction of time-frequency features also called as spectrogram. Energy, variance, and skewness features are computed on the spectrogram. These features are used to train a support vector machine and a neural network classifier. The classifiers are tested for person authentication with testing data using cross-validation. Results using a different number of channels with optimum features are presented. A Graphical User Interface is also presented for easy use of the person authentication system.