The Research And Implementation Of 3D Modeling And Expression Generation Algorithm For Face Image

With the development and popularization of 3D film and television entertainment works,people's pursuit of images has also increased from two-dimensional to threedimensional.3D works are becoming more and more popular,and the threedimensional face is the most concerned information in all kinds of works.Although 3D face data can be obtained through device scanning,it cannot be obtained from existing face images,so 3D face reconstruction based on a single face image is of great research significance.In the reconstruction process,most of the face images used as input are not frontal and good,so it is necessary to complete the missing parts of the face in the input image during the three-dimensional reconstruction.At the same time,in the application of three-dimensional human faces,it is necessary to make it produce expressions to represent facial movements,so the generation of facial expressions also has research significance.This dissertation takes the key technology of 3D face reconstruction and expression generation as the main research object,and makes improvements to the surface shape reconstruction of 3D faces and complements the surface texture,and generates facial expressions to achieve the effect of deformation.The main research contents of this dissertation are as follows:(1)For the end-to-end method of rebuilding a 3D face,there is no unified 3D face representation method and the problem of insufficient face feature extraction.This dissertation adopts an improved UV position map method and uses Dense Net to extract the input face image.Features to complete the reconstruction of the three-dimensional face surface shape.Experimental results show that the algorithm in this dissertation can effectively improve the accuracy of 3D face shape reconstruction,and the improvement effect will be better for faces with larger angles as input.(2)Aiming at the problem that the input face image information is insufficient to generate a complete texture,this dissertation proposes to use the method of face frontalization to fill in the missing texture,and proposes a self-supervised method to solve the lack of face frontalization research Problems with training data.The experimental results show that the algorithm in this dissertation can better supplement the missing texture information,and it has a good effect even when compared with the algorithm of face frontalization.(3)In view of the insufficient use of facial features by the current facial expression generation algorithm and the distortion of the generated face shape,this dissertation proposes a latent space-based method to generate facial expression images and obtain them through motion estimation.Complete and continuous facial expressions.The experimental results show that the algorithm in this dissertation can generate better facial expression images and preserve the facial structure better.Finally,all the experiments are integrated to form a three-dimensional face surface reconstruction system,and the above experiments are jointly verified,which proves the validity of the experiment and the integrity of the system.At the same time,the overall error of the system is tested and analyzed.