Regional registration and curvature descriptors for expression resistant 3D face recognition

Registration plays a vital role in 3D face recognition. In the registration phase, some fiducial points are needed. In this paper, a method is proposed to automatically localize landmark points. In addition, the performance of registration is affected by deformations caused by expression variations. It is asserted that regional registration and recognition is resistant to facial deformations. Finally, we propose to use translation and rotation invariant curvature descriptors as surface features to deal with erronous registration. In experiments conducted on the Bosphorus 3D face database, which contains significant expression variations, the performance of the proposed system was satisfactory.