Fourier-Bessel domain based band-limited entropies for automated detection of human emotions

Abstract

In various aspects of our daily lives, emotions play an important role in behavior, decision making, cognitive learning, perception and rational thinking. Therefore, analyzing emotions is key to understand human nature. Emotions can be recorded using facial expressions, galvanic skin response, speech signals, electroencephalogram (EEG) signals, etc., In this work, we use EEG signals for emotion detection for its advantages over the rest of the approaches especially in terms of signals getting changed when the person tries to hide his/her emotion. An EEG based dataset is used to validate the proposed approach. The emotions are evoked by showing the subjects videos which stimulate happy, sad and neutral emotions. We propose a Fourier-Bessel (FB) domain based band limited entropies for classifying emotions. We introduce a novel concept of band-limited entropies which is computed for each sub band without decomposing the signal completely. Once the entropies are obtained, they are used as features for classification. A few machine learning classifiers such as support vector machine (SVM), k-nearest neighbor (KNN) and their variants are used to perform classification. It is observed that fine KNN classifier has performed well with an accuracy of 92.3% with less computational complexity.