In order to increase the accuracy of motor imagery signal detection in a brain computer interface system, choosing the right features is a key point. There are several methods for brain signal feature extraction. The most commonly used method uses features from the frequency-domain. The classification accuracy rate achieved by extracting features from the frequency-domain is quite good, but extracting particular features in the time-domain is still being explored as an alternative for getting better accuracy.

Cross-correlation is a method for measuring the similarity between two signals and results in a cross-correlation sequence. Basic statistic parameters can be taken from the cross-correlation sequence as features of a signal and then used in a classification process. To achieve a high classification accuracy rate, we should choose the appropriate signal as the reference signal when applying cross-correlation and also choose which basic statistic parameters to take as features. Model validation can be used for evaluating the reference signal and the basic statistic parameters in relation to the accuracy rate.

The method proposed here for choosing the reference signal and the basic statistic parameters was applied and tested with the BCI Competition III, IVa dataset. Using 10-fold cross validation for model validation, the proposed method obtained a classification accuracy rate of 99%. This rate was attained by using the signal from channel CFC2 as the reference signal using the maximum value from the cross-correlation sequence as the only feature. This result is approximately 4% better than the accuracy achieved in a previous study that used a similar method. Not only the accuracy rate was improved in this study, but it was also shown that the maximum value of the cross-correlation sequence could potentially be used as sole feature.

brain computer interface (BCI), electroencephalograph (EEG), fast Fourier transform (FFT), cross-correlation, reference signal.