Characterization of Emotions Through Facial Electromyogram Signals

Abstract

Emotions are a high interesting subject for the development of areas such as health and education. As a result, methods that allow their understanding, characterization, and classification have been under the attention in recent years. The main objective of this work is to investigate the feasibility of characterizing emotions from facial electromyogram (EMG) signals. For that, we rely on the EMG signals, from the frontal and zygomatic muscles, collected on 37 participants while emotional conditions were induced by visual content, namely fear, joy, or neutral. Using only the entropy of the EMG signals, from the frontal and zygomatic muscles, we can distinguish, respectively, neutral and joy conditions for 70% and 84% of the participants, fear and joy conditions for 81% and 92% of the participants and neutral, and fear conditions for 65% and 70% of the participants. These results show that opposite emotional conditions are easier to distinguish through the information of EMG signals. Moreover, we can also conclude that the information from the zygomatic muscle allowed to characterized more participants with respect to the 3 emotional conditions induced. The characterization of emotions through EMG signals opens the possibility for a classification system for emotion classification relying only on EMG information. This has the advantages of micro-expressions detection, signal constant collection, and no need to acquire face images. This work is a first step towards the automatic classification of emotions based solely on facial EMG.