Research On Face Semantic Parametric Model And Its Application

3D face model editing is one of the important means of face modeling.De-signers often use manual work to edit existing face models into ideal face shapes.However,this editing mode does not apply to ordinary users.For the ordinary user,the simplest and most effective way is to intuitively edit the face shape by controlling low-dimensional parameters.As the most popular face parameterized bilinear model at present,the face is expressed as a linear combination of the face shape space base and the expression space base.The weight coefficients of these bases are shape parameters(identity)and expression parameters(emo-tion).Because the expression base has semantics,the expression parameters can be intuitively adjusted to obtain specific facial expressions(happy,anger,etc.).However,it is not possible to obtain a specific human face shape by adjusting the face shape parameters(high nose,large eyes,etc.).In order to satisfy the face shape editing according to the user's wishes,how to use the face parametric model to have semantic control of the face shape is the focus of this paper.In this paper,the facial parametric model is deeply studied.By studying how to use parametric models to intuitively control the face identity shape,a semantic facial model is proposed.Based on the semantic model,the paper explores the potential application value of face editing,and has achieved certain progress in the facial image editing and the customization of users' digital avatars.The following specific research results have been achieved:· In terms of the problem of the lack of semantic parametric control on face shape in the existing face parametric model,a face semantic parametric model is proposed.By introducing a semantic base vector,satisfying the user's desire to directly adjust specific semantic parameters can obtain prac-tical facial deformation effects(such as raising the bridge of the nose,bigger the eyes,etc.)to provide users with more intuitive and easy means to oper-ate.In addition,the semantic base has the local control ability of the human face shape,constructs a space with a rich face shape,and better displays the data generalization ability.The semantic base of the semantic local de-formation of the template face structure is constructed.Then the semantic base is used as a constraint,and the potential face shape space learning without semantics is discovered through a data-driven method.Qualitative and quantitative experimental analysis shows that the face semantic para-metric model presented in this paper exhibits higher performance for 3D face reconstruction and 3D face semantics editing.· In terms of the problem that the face shape editing method in images has strict requirements on the face pose in the selected image,a face image editing method based on a face semantic parametric model is proposed.This method not only satisfies the user's need to semantically edit the face shape,but also satisfies the user's requirement for replacing face parts be-tween images.The method combines face image segmentation,face model reconstruction,mesh editing,and image fusion techniques,to represent hu-man face in 3D space by using a human face semantic parametric model,and then performs face shape editing and final rendering face image.The experimental results show that this method has richer deformation options for face shape editing than those image editing methods in image space,and there is no restriction on human face pose in the image.· In terms of the application requirement of the digital avatar for ordinary consumer,a personalized digital avatar construction method based on face semantic parametric model is proposed.This method requires only three images of the user(two for the whole body and one for the face),and the 3D model of the human body and the face reconstructed from the image is fused using the mesh fusion technique to a digital avatar with facial geometric information.In terms of the problem of blending boundary smoothing of the avatar,this work proposes an efficient triangle mesh quality optimization method,which ensures the smooth visual effect of the user digital avatar.The introduction of face-semantic parametric model not only enhances the face model's geometric features,but also allows the user to semantically edit the face shape to obtain a personalized avatar.The avatar model constructed by this method can be applied to 3D games virtual fitting,3D printing,etc.,and has a wide range of practical value.