Learning Encoded Facial Curvature Information for 3D Facial Emotion Recognition

In current 3D facial expression recognition system, feature extraction has always been a critical point. We focus on encoding feature by using curvature information. 3D facial expression images are described by means of four images which gray level are the value of curvature-based descriptors (principal curvatures k₁, k₂, mean curvature, shape index) and then encoded by LBP. SVM classifier is employed for classification. Then experimental result illustrates that the proposed feature classified by SVM is effective.