3D Morphable Model Based Face Reconstruction And Its Application

Face is always a popular area in computer vision and computer graphics.With the popularity of smart phones,users have huge demand for face related tasks such as face beautification,face recognition,facial animation and so on,which encourage more researchers pay attention to face related works.Meanwhile,a large number of face images on social networks also benifit the research.Some of the existing research,such as face recognition,face alignment,have been caught in a bottleneck only rely on 2D information,because the face pose and illumination are various in real world.Light and pose condition should be modeled in 3D domain,to improve the accuracy of face recognition and face alignment.However,3D reconstruction from 2D images is a ill posed problem.In order to solve this ill posed problem,some prior knowledge have been introduced.3D morphable model is one of the most common ones.It is a statistical model,which is the extraction and simplification of the real 3D face model.In this paper,we propose a fast and efficient 3D face reconstruction method based on 3D morphable model.Instead of traditional analysis-by-synthesis method based on 3D morphable model,our method is landmark-driven.We establish a mapping function from dense 3D model to sparse 2D facial landmarks.Only relying on facial landmarks will lead to overfitting problem,Therefore,we exploit the texture information near the landmark,and used discriminable local features to guide the updating of model parameters.Experiments show that our algorithm is very fast and robust.Each variable in our model has its corresponding physical meaning.Therefore,once we solve these parameters,our algorithm can also be applied to face alignment,face tracking,face pose estimation,face animation and other fields.