Recognition of Emotion Through Facial Expressions Using EMG Signal

Emotion recognition play important role in human-computer interfacing and a treatment of a person under depression. Facial expressions of a person reflect his emotional status. Electromyogram (EMG)based emotion recognition systems able to recognize true emotions of a person. Current research on EMG based emotion recognition reports overall accuracy in the range 69% to 91 % in a particular emotional environment. In case of posed expressions, emotions were recognized with accuracy range from 91 % to 97%. There is a scope for improvement for enhancing accuracy of emotion recognition in emotional environment. In this research work EMG dataset acquired under emotional environment by Augsburg University is analyzed. From 96 EMG signals representing four emotions, four features including Root mean square, Variance, Mean absolute value and Integrated EMG are calculated. These parameters are given to 3 different classifier namely Elman neural network (ENN) classifier, Back propagation neural network (BPNN), and Nonlinear autoregressive exogenous network (NARX) for classification of emotion. NARX neural network gave maximum overall accuracy of 99.1 %.