Analysis and Comparison of the Cone Curvature Descriptor in Facial Gesture Recognition Tasks

This article presents the results of analyzing the behavior of the Cone Curvature shape descriptor (CC) in the task of recognition of facial expressions in 3D images. The CC descriptor is a representation of the 3D model computed from a set of waves modeling for each vertex of a polygon mesh. The 3D Facial Expression Database (BU-3DFE) was used, which contains images with six facial expressions. With the use of the CC descriptor, the expressions were recognized in an average percentage of 76.67% with a neural network, and of 78.88% with a Bayesian classifier. By combining the CC descriptor with other descriptors such as DESIRE and Spherical Spin Image, it was achieved an average percentage of gesture recognition of 90.27%and 97.2 %, using the mentioned classifiers