3D face recognition based on G-H shape variation

Face recognition has been an interesting issue in pattern recognition over the past few decades. In this paper, we propose a new method for face recognition using 3D information. During preprocessing, the scanned 3D point clouds are first registered together, and at the same time, the regular meshes are generated. Then the novel shape variation representation based on Gaussian-Hermite moments (GH-SVI) is proposed to characterize an individual. Experimental results on the 3D face database 3DPEF, with complex pose and expression variations, and 3D_RMA, likely the largest 3D face database currently available, demonstrate that the proposed features are very important to characterize an individual.