Automatic Human Facial Affect Classification Using Computational Intelligence Techniques

Automatic and robust emotion recognition is of prime concern while designing affect-sensitive machines for human-computer interaction to fully understand the emotional context implicated. This paper presents the development of a Decision Tree (DT) based scheme for the automatic classification of elementary human facial emotions. In order to implement affect classification algorithms, this research employs the Cohn-Kanade dataset based on facial features. Action units are mapped to emotion labels using the Facial Action Coding System (FACS); these labels serve as targets for supervised learning. The proposed scheme is validated through simulation, and results are compared to the existing techniques like Back Propagation Neural Network (BPNN) and Support Vector Machine (SVM). The presented results substantiate that the classification scheme suggested in this paper performs extensively superior than other alternatives in terms of accuracy with minimal computational efforts. Hence, DT based affect classification algorithm is an excellent alternative for applications comprising emotion recognition.