Research On Fine-grained 3D Face Reconstruction From A Single Image

3D face reconstruction has a wide range of applications in many fields such as digital entertainment and virtual reality.It has long been widely concerned by computer vision researchers.At present,high-precision 3D face reconstruction technology based on complex equipment and controllable environment has been successfully achieved in 3D film production.However,with the popularity of low-cost smart phone and the growing demand for entertainment and social networking,There is an urgent need for fast and accurate 3D face reconstruction based on simple unconstrained images.Due to lack of information such as depth information,camera parameters and lighting environment,it is often impossible to achieve satisfactory results when dealing with single image.In response to this problem,this paper makes full use of a priori information including 3D face morphable model and illumination prior distribution.An algorithm for reconstructing a fine 3D face model from a single image based on staged optimization is proposed.In this paper,based on the face feature point correspondence between image and deformation model,we propose two algorithm to estimate face poses,face appearances and face expression parameters.For different feature point detection methods,we use the principle of weak perspective projection and rigid transformation to estimate the parameters respectively.After the iterative solution of the algorithm,we will generate an initial 3D face model that fits into the input face image.It will provide a reliable template for subsequent facial intrinsic characteristics research and refined face model reconstruction.Then,the paper studies the intrinsic properties and illumination environment of face images.We assume the face as a Lambertian object containing only diffuse reflections,and thus use parametric texture models and spherical harmonic illumination to characterize the reflectivity properties and the lighting environment of the face.Combining the a priori constraints imposed by the face lighting environment data set and the known parametric initial 3D model,we implement the intrinsic decomposition of the face image in three-dimensional space,and extract the corresponding face texture and lighting environment information,which will provide more auxiliary information for further refined face reconstruction.Finally,this paper proposes a face geometry detail optimization algorithm based on image shading information.We construct the constraints that the target fined 3D face model should satisfy from some perspectives,such as reconstruct image consistency,shading change consistency,overall face consistency,and gradient integrativity.Combining the auxiliary information of the previous work,we first optimize the gradient field of the surface height field of the model,and then intergrate to the height field.After model vertices triangulation,we eventually achieve the reconstruction of a fine 3D face model with geometric details.The general algorithm proposed in this paper combines the advantages of the morphable model method and the shape-from-shading method while avoiding the respective shortcomings.The morphable model-based algorithm provides a good overall facial shape,while the intrinsic property decomposition of the face image makes the shape-from-shading become feasible.And finally the method of shape-from-shading makes the reconstruction result more refined and personalized.The experiments in this paper show that the 3D face reconstruction algorithm proposed in this paper outperforms some existing methods in terms of accuracy and visual perception.