3D measures and geometric facial features analysis on 3D facial action unit recognition

We constructed a statistical model for variations in facial shape due to differences in 22 facial action units (AUs) using three different 3D measures and geometric facial features as the feature vectors. The three facial features are 3D facial points, 3D facial surface normals and 3D distance measurements. Finally, we apply Support Vector Machines as the classifier and 10-fold cross validation are performed on the Bosphorus database. Using our approach, 3D facial points yields a better performance than 3D facial distance measurements and 3D facial surface normal in 3D Facial AU recognition