Expression-insensitive 3D face recognition by the fusion of multiple subject-specific curves.