Cepstrum Based Algorithm for Motor Imagery Classification

Abstract

A linear convolutive mixing model based real time motor imagery classification algorithm is proposed in this paper. The proposed cepstrum based method is very first and robust unsupervised learning algorithm, extremely useful for real time brain computer interface(BCI). The cepstrum is analyzed for estimation of combined action potential generated through the active synapses of raw electroencephalogram (EEG) signal. Maximum energy of the estimated cepstrum, is used as a feature. The extracted feature further subjected to simple Bayesian probabilistic classifier, for classification. The proposed method of EEG signal pre-processing and feature extraction outperforms the conventional temporal relative spectral power (TRSP) based movement imagery classification algorithm and BCI competition II results.