Emotion Recognition from Facial Expression using Explainable Deep Learning

Facial expressions give important information about people emotions. Understanding facial expressions accurately is one of the challenging tasks for interpersonal relationships. Automatic emotion recognition using facial expressions is an area of interest within various fields as, for instance, computer science, healthcare, bio-engineering and psychology. In this paper we propose a method for automatic emotional recognition from facial expressions by exploiting deep learning techniques. The proposed method is able to classify a facial image under analysis in one of three different emotions: happiness, neutral or sadness. Moreover the proposed method is able to show on the facial images the areas symptomatic of a certain prediction, thus highlighting the characteristics of the face that are related to a particular emotion: in this way the model is able to explain the reason why it predicts a certain emotion. A preliminary experimental analysis performed on a 1,500 different facial expression images shows that the proposed method is promising in emotion recognition.