3D Face Reconstruction with Geometry Details from a Single Image

摘要

3D face reconstruction from a single image is a classical and challengingproblem, with wide applications in many areas. Inspired by recent works in faceanimation from RGB-D or monocular video inputs, we develop a novel method forreconstructing 3D faces from unconstrained 2D images, using a coarse-to-fineoptimization strategy. First, a smooth coarse 3D face is generated from anexample-based bilinear face model, by aligning the projection of 3D facelandmarks with 2D landmarks detected from the input image. Afterwards, usingglobal corrective deformation fields, the coarse 3D face is refined usingphotometric consistency constraints, resulting in a medium face shape. Finally,a shape-from-shading method is applied on the medium face to recover finegeometric details. Our method outperforms state-of-the-art approaches in termsof accuracy and detail recovery, which is demonstrated in extensive experimentsusing real world models and publicly available datasets.