Compound Facial Emotional Expression Recognition using CNN Deep Features

Most advanced researches concerning facial expressions identification generally deals with the six basic emotions with a single tag, namely happiness, anger, disgust, fear, surprise and sadness in addition to the neutral facial expression. It has recently been discovered that some affective computing analysts are more interested in delicate emotions such as a compound facial expression that deals with primary and complementary emotions (e.g. "happily-surprised" and "sadly-angry"). In addition, these emotions are more complete than the seven basic facial emotions. Therefore, current research focuses on the examination of fundamental and compound emotions, using databases of Compound Facial Expressions of Emotions (CFEE) and spontaneous Real-world Affective Faces (RAF). In this article, transfer learning based on Residual Neural Network was experimented to create a deep feature extraction model, and then different machine learning methods are used for classification. Experiments on basic and compound emotions prove that the proposed system can perfectly surpass advanced approaches. As expected, the compound emotions are more difficult to identify than the seven basic emotions.