Comparison of Feature Extraction Techniques for EEG Based Brain-Computer Interface

The analysis of electroencephalogram(EEG) signals, for implementation of brain-computer interface (BCI), has enticed a lot of interest in the research community. It can be used in a variety of applications ranging from medical rehabilitation to pleasure and entertainment. BCI is very promising research domain even in the face of a number of challenges, especially in domain of signal processing, feature extraction and that of classification techniques, as these EEG signals contain considerable amount of noise and artifacts. In this paper, various feature extraction algorithms used in BCI are investigated and compared. Common spatial pattern(CSP) is a popularly applied algorithm for extracting features from EEG signals in an implementation of BCI. Filter bank spatial pattern (FBCSP) and spectrally weighted common spatial pattern (SWCSP) are further extensions of CSP. The performance of these methods is evaluated and calculated on data set 2a of BCI competition IV, on the basis of standardized mean squared error (SMSE) . Results shows that FBCSP considerably outperforms the performance of the other methods under consideration