Research On High-precision 3D Face Registration Method

High-quality 3D face model reconstruction technology have broad application and development application prospects in film and television production,3D animation and other fields.Obtaining high-precision 3D face models is of great significance to the development of these fields.For the standardization and highefficiency development and application of 3D faces,the face models need to have the same topological structure,so 3D face registration technology needs to be used,and respectively use the same standard template model to represent all face models with same topological structures obtained by scanning Face model.so as to achieve the topology uniformed.The traditional face registration method has the problem of low registration accuracy in the high-frequency region of the face.In order to further solve this problem,this paper adopts the deep learning method based on graph convolutional neural network to improve the original reconstruction model on the basis of the traditional method.For precision registration,a bilinear parameterized model is constructed based on the registered model,and finally a new 3d face model is reconstructed by fitting the bilinear model.The main work is as follows:1.Build a high-resolution multi-view face image synchronization acquisition system to perform three-dimensional reconstruction on the collected multiview face images to obtain a high-precision three-dimensional face model.Then use the nearest neighbor iterative algorithm to perform preliminary registration on the reconstructed 3D face model,and repair the symmetry of the face model grid after the preliminary registration,and build a high-precision 3D face database;2.Propose a deep neural network based on the basis of graph convolutional neural network,and perform network training based on the constructed 3D face data-set to achieve the output from the input Raw Scan model to the topologically uniformed face model.3.The above-mentioned face model after neural network registration is used to construct a bilinear parameterized face model,and the parameterized model is used to perform multi-viewpoint fitting to generate a new 3D face model.