Optimal Parameterization Selection for the Brain-Computer Interface

The contribution deals with the optimization of the EEG off-line, single-trial movement classification by means of parameterization tuning. The data we classify represent manifestations of the simple movements performed by the right shoulder (proximal movement) and right index finger (distal movement) of experimental subjects. We implemented several approaches to the EEG parameterization and compared results in order to increase the recognition score. The results are compared with the results from our earlier works and will form a strong basis for the coming experiments with a new EEG database. The target of our experiments is the implementation of the Brain Computer Interface machine recognizing movements performed on one side of the body using the non-invasive EEG scanning. Key-Words: Brain-Computer Interface, finger movement, shoulder movement, parameterization, classification, hidden Markov models