High-Quality 3D Face Reconstruction from Multi-View Images

This paper presents a method to reconstruct high-quality 3D face model from multi-view images with unknown poses,expressions,and illumination conditions.We adopt a multi-stage based optimization method to solve this challenging problem.Specifically,we first fit the input multi-view images with a parametric model by an analysis-by-synthesis method.Next,dense correspondence between input facial images is constructed by solving an optical flow problem on the albedo domain,and a point cloud with person-specified geometry characteristics is then recovered based on the reliable dense correspondence.In the final,fine-scale details are reconstructed by using a multi-view shape from shading method.Experiments on synthetic and real data demonstrate that our proposed method is able to reconstruct accurate 3D face models with fine geometric details,and the quantification studies show that our method is better than state-of-the-art methods.