Data-driven Facial Animation Based On Hypergraph Learning

With the rapid development of computer technology,Computer Graphics(CG)has been increasingly applied to the digital movies and games.Among them,facial animation is one of the most typical research fields of Computer Graphics.In recent years,data-driven facial animation has caught many researchers' attention.So far,the existing facial animation driven methods obtain the driving results simply by linear transforming the given data.However,these methods have their own inherent drawbacks: firstly,the driving results depend on the selected key models;secondly,their computation complexity is high and this process is time-consuming;finally,these methods cannot achieve a natural face model.During the data-driven facial animation process,the source face should keep the same topology structure with the target face.And the topological structure of the face model can be described by the relationship between the vertices on the face model.Therefore,this paper proposes a new approach to solve the problem of high order relationship vertices based on hypergraph learning,and proposes a new data–driven facial animation method.In this method,we first construct a hypergraph using all the 3D vertices on the face model;secondly,we assign the weight of each hyperedge according to the similarity between the vertices,and then calculate the Laplacian matrix of the hypergraph;finally,we use the Laplacian matrix to realign the vertices,and obtain the driving face model.In addition,in order to obtain more realistic and natural driving results,we need to select some vertices from the target face model as a constraint.Some of the representative vertices are chosen manually and the rest of vertices are chosen randomly.The experimental results show that our method not only maintain the topology structure between the source face model and the target face model,reduce manual intervention,and improve the audience's visual experience,but also obtain a more accurate and natural facial model.It also shows the superiority over the state-of-art.