Automatic and accurate 3D face registration under the guidance of intra-class difference measurement

The accuracy of 3D face registration algorithm greatly influences the effect of 3D face recognition. While a lot of efforts have been dedicated in developing automatic and accurate registration methods, these methods usually cannot be compared and evaluated on a fair basis because of the lack of a standard quantitative measurement. In this paper, we propose a practical quantitative analysis method based on intra-class difference to evaluate the accuracy of face registration methods, and apply it to guide the procedures of an automatic nose symmetry plane (NSP) method. After every step, we calculate the mean and standard deviation (STD) of intra-class pose differences for all involved face images to assess the effect of this step and determine how to further improve the accuracy in the following steps. Extensive experiments have been conducted using the FRGC (V1.0 and V2.0) benchmark 3D face dataset to demonstrate the feasibility of our guided registration method.