A novel integral and SE(3)-invariant description for 3D face recognition

We propose a novel description of 3D faces for the task of face recognition. It is based on an integral approach to avoid the errors caused in the case of derivative methods. The geodesic distance computing, which is considered as an integral approach, is performed. The geodesic three-polar parameterization is, firstly, implemented on 3D faces to remove the dependence with regard to the original mesh which is known to be a difficult problem in the 3D data analysis context. Then, the geodesic distances between the pairwise points of the three-polar parameterization are computed to form the Geodesic Distance Matrix description. Intensive experimentations are performed on the BU-3DFE and the Bosphorus face databases. The obtained results showed the accuracy of the proposed integral method for the face description relatively to the identification and the verification protocols. Very competitive rates with the state of the art methods are, also, obtained.