Classification of emotional states through the investigation of brain activities

Abstract

Emotion one feels is generally expressed via face and speech. In this paper, it is investigated whether it is possible to recognize emotional states through EEG signals produced while the brain performs some cognitive function or not. Up to now, significant efforts have been made mainly by psychologists that try to define various emotional states based on EEG data acquired during the presentation of stimuli. But judging from the engineering point of view, these attempts have many limitations that only a few features such as power and asymmetry score have been used to elucidate the characteristics of emotions, and application of the found results to the Human-Computer Interface (HCI) field has not been considered. Thus, to overcome these limitations and weakness, empirical knowledge from the fields of psychology and music therapy, and various methodologies on the signal processing are introduced simultaneously in this study. Fist, affective stimuli consisting of images, music and video clips are designed and prepared for EEG recordings. This work is inevitable for gathering good and reliable data. Second, in the signal analysis phase, an endeavor to find the effective features reflecting information inherent in EEG signals is made, and an approach to select the optimal channels based on the common Principal Component Analysis (PCA) is also made. In the classification test, it was found that the 4 emotional states such as relaxation, happiness, sadness, and stress could be distinguished if the information gathered from the brain activities occurring in the frontal and temporal regions was utilized.