Fully automatic 3D facial expression recognition using differential mean curvature maps and histograms of oriented gradients

In this paper, we propose an holistic, fully automatic approach to 3D Facial Expression Recognition (FER). A novel facial representation, namely Differential Mean Curvature Maps (DMCMs), is proposed to capture both global and local facial surface deformations which typically occur during facial expressions. These DMCMs are directly extracted from 3D depth images, by calculating the mean curvatures thanks to an integral computation. To account for facial morphology variations, they are further normalized through an aspect ratio deformation. Finally, Histograms of Oriented Gradients (HOG) are applied to regions of these normalized DMCMs and allow for the generation of facial features that can be fed to the widely used Multiclass-SVM classification algorithm. Using the protocol proposed by Gong et al. [1] on the BU-3DFE dataset, the proposed approach displays competitive performance while staying entirely automatic.