Physical Based Inverse Rendering of 3D Faces from a Single Image

In this paper, we present an automatic framework for generating 3D faces in real-time VR systems or games from a single image. The framework are divided into reconstruction stage and rendering stage. In the first stage, we propose a linear machine learning model which is faster than previous works because features of illumination are learned from images in advance. In the second stage, we propose a physical based method to generate skin details and an approximation rendering algorithm. Finally, our work is verified in a realtime graphic application where we can view the face model with strong realism and incredible resemblance to the origin face photo.