N2 components as features for brain computer interface

Abstract

A mental speller using brain computer interface (BCI) may allow a user to communicate by gazing at a virtual keyboard on the screen to select a desired character to compose a word, and thus sentences. Different from other paradigms, a so called imitating-natural-reading (INR) modality was exploited to construct a novel mental speller-INR SPELLER. In order to boost the bit rate, a 300ms window was used to estimate the accurate time of target stimuli onset from EEG signals. To meet this task, N2 components of visual evoked potentials (VEP) were investigated. Experimental results indicated that the object-specified component can be estimated in single trial at an accuracy of 90.5% with support vector machine (SVM) classifier.