Correspondences between expressive 3D faces based on radial basis function re-sampling

This paper presents a method that can directly establish dense correspondences between 3D faces with arbitrary expressions and identities. We propose a re-sampling algorithm which uses a reference 3D face to re-sample target 3D faces under the guidance of radial basis function (RBF) network. The algorithm not only establishes dense correspondences between 3D faces, but also interpolates the regions in a target 3D face where 3D data are absent for the reason of range data acquisition. We also address the problem of artificial effects in mouth region caused by the prime RBF. By examining the essence of RBF, we define a unique distance metric for lips to solve the problem. In addition, the feature point detection procedure is also well designed to alleviate manual interaction. The experimental results show that our method is effective for correspondence computation of 3D face data with even exaggerated expressions.