3D Face Reconstruction Based On A Single Rgb Photo

With the rapid development of science and technology,people begin to pursue higher quality of life and enjoyment.They are no longer satisfied with the sense experience of plane space,but start to pursue the three-dimensional visual enjoyment.Three-dimensional face reconstruction is a hot research issue in the field of computer vision.At present,the most common methods of 3D face reconstruction are depth based image reconstruction,multi vision based reconstruction and statistical data based reconstruction.The data acquisition equipment of the first two methods has a high cost and requires a fixed scene,which requires users to interact with the equipment in a short distance and maintain a fixed attitude.It can not realize remote 3D face reconstruction,which not only costs a lot,but also has a complex process,and can not meet the needs of a large number of users.The third method can be based on a single RGB image reconstruction,which can avoid the above problems.In addition,the face image has diversity and uncertainty.A single face image is collected from a single perspective by the camera.Part of the face information cannot be collected due to occlusion,which will lead to the lack of texture.Therefore,in view of the above problems,the main research contents and proposed methods are as follows:(1)A method of 3D face reconstruction based on a single face image.To solve the problem that a single RGB image can't collect complete face depth data,based on the 3D deformation model,this paper adopts the cascade regression method based on convolution neural network to input the original face image after clipping and the projection image after projection normalization and Z-buffer processing,and outputs the parameters of 3dmm model to reconstruct 3D face,which is affected by the database model Because of the limitation and low shape matching,UV location map is introduced to represent 3D face,and the 3D face is reconstructed end-to-end by coding and decoding network.(2)A complete texture generation method based on a single face image.To solve the problem of texture missing in 3D face reconstruction based on a single image,this paper proposes a texture restoration algorithm based on two-way face mapping.Firstly,a single face image is generated by CR-GAN to generate a multi angle face image and 3D reconstruction is carried out,3D face is aligned with the image,and initial 3D face texture and 3D face mask are obtained by vertical sampling.By repairing the 2D mapping image obtained by 3D texture mesh parameterization,and using Poisson fusion and other optimization methods,a complete texture map with realistic sense is generated.(3)Face migration algorithm in digital modeling design based on 3D face pose adaptation.In this paper,we design a digital modeling design method based on face migration,and propose a large angle face migration algorithm to solve the problem of edge transition and face pose mismatch in traditional face migration.Three dimensional face reconstruction technology and face deformation technology are applied to face migration.The face shape is changed by using triangle mesh affine transformation,and the fusion face is preprocessed by three-dimensional pose correction and skin color template migration,and then the face migration is carried out to get a more natural and real face fusion effect,so as to achieve the purpose of changing the shape.