3D Human Face Surface Reconstruction from Unstructured Point Cloud Face Images

Abstract

Application such as a 3D face scan fitting to a 3D face model required 3D face in mesh triangulation representation for the application to work. The 3D face model can do augmentation such as variety of facial expression and poses which enlarge a 3D face dataset. 3D face unstructured point cloud presentation needs to go a certain process to convert it to 3D face mesh consist of faces, edges and vertices. In this paper, it evaluated a methods, such as screened Poisson surface reconstruction, Poisson surface reconstruction, an extension of Poisson surface reconstruction with removing vertices with low density, ball pivoting surface reconstruction, and Shape As Points (SAP) on the Bosporus dataset. In addition, to achieve results with little imperfection, preprocess steps are implemented before surface reconstruction. The result shows triangulated mesh differ in visualization between each method.