Fuzzy SVM for 3D Facial Expression Classification using Sequential Forward Feature Selection

Facial expression detection is one of the emerging topics in computer vision. In this study, three-dimensional (3D) facial expression classification has been addressed. Firstly, a large set of features based on pair-wise distances of points in face model are extracted. The multi-class problem of facial expression detection is divided into 15 one-versus-one two-class classifiers. Sequential forward feature selection (SFFS) algorithm based on Naive Bayesian error rate is applied to select the most discriminative features. In the last step, a two level fuzzy SVM (FSVM) classifier is utilized in optimum low dimensional feature space to detect multi-class labels of six basic expressions including anger, disgust, fear, happiness, surprise and sadness. Experiments conducted on BU-3DFE data set have proved that the performance of proposed algorithm is comparable with recent studies in this field.