Single-trial Classification of Viewed Characters using Single-channel EEG Waveforms

Abstract

Electroencephalograms (EEGs) and Eventrelated potentials (ERP) have long been used to observe the human visual perception process, such as the human response to letters, Kanji characters and symbols. This paper examines the possibility of classifying characters when viewed by subjects in single trials using single-channel EEG waveforms of the frontal area (Fz) and the occipital area (Oz) of the brain. The first 20 trials for each character were used for calibration, and the remaining trials were assigned to the test data set. Feature vectors for each trial were created as EEG waveforms from 100 up to 800 msec. after the stimuli was shown. To extract features of waveforms, the regression relationship between EEG and ERP waveforms was used to transform observed signals. As a result, the performance of cross validation rates of the test data set increased incrementally during the perceptual process, for both Fz and Oz, when the predicted waveforms were measured using the regression relationship. Also, the effectiveness of the prediction using the regression relationship for the classification performance of viewed characters was determined during the perceptual process. This provides evidence that a procedure using the relationship between EEG and ERP is effective in predicting viewed characters.