Adapting Discriminative Spatial Filter Based on Movement Related Potentials for BCI

Focusing on the unclear distribution of the recorded electroencephalogram (EEG) data and the relative inadequate of training data in brain-computer interface (BCI), a novel approach, called as adapting discriminative spatial filter (ADSF), is proposed for the extraction of movement related potentials (MRPs) features. ADSF searches for the optional direction, which maximizes the ratio of the local between-class distance and the local within-class distance in the projected space. The distances defined require only information about the neighborhood, while no assumption is made about the underlying data distribution. ADSF adaptively measures the neighborhood relation and estimates the similar probability of pairs of samples more reasonably. At last ADSF was applied to two datasets from BCI competition 2001 and 2003. The experimental results validate the efficiency of ADSF.