Abstract:

In order to remedy the complex distribution of recorded electroencephalogram （EEG） data and the shortage of training data in terms of brain-computer interface （ BCI）, a novel approach named neighborhood spatial pattern （NSP） is proposed to extract movement-related potentials （MRPs）, which constitute the most important fea- tures utilized in the classification algorithms for the motor-imagery-based BCI. NSP searches the optimal direction which maximizes the ratio of the between-class distance to the within-class distance of the neighborhood in the pro- jected space. During the search, no assumptions about the latent data distribution should be made, and only the neighborhood relationship and the label information are required. NSP is also applied to two datasets from BCI com- petitions 2003 and 2001. The results show that NSP can effectively extract MPRs features.

Key words: neighborhood, feature extraction, brain-computer interface, movement-related potential, event-relateddesynchronization/synchronization