Classification of EEG Signals Using Filter Bank Common Spatial Pattern Based on Fisher and Laplacian Criteria

Common spatial pattern (CSP) is a highly successful algorithm in motor imagery based brain-computer interfaces (BCIs). The performance of the algorithm, however, depends largely on the operational frequency bands. To address the problem, a filter bank was applied to find optimal frequency bands. In filter bank, CSP was applied in all sub-band signals for feature extraction. The feature selection is the key of filter bank method for increasing classification performance. In this study, coefficient decimation (CD) technique was used to devise filter bank, while Fisher score and Laplacian score were proposed as feature selection criterion. In off-line analysis, the proposed method yielded relatively better cross-validation classification accuracies.