Research On 3D Face Reconstruction Algorithm Based On Self-occlusion

3D face reconstruction technology has long been a very active field of computer graphics and human-computer interaction.With the maturity of hardware and software,the application of 3D reconstruction technology is becoming more and more in the fields of film,game,security and map.Multi-dimensional,and thus three-dimensional modeling has also been more and more people concerned about the three-dimensional face model reconstruction is one of the key technologies in the field of 3D modeling.At present,the methods of obtaining three-dimensional model mainly include three kinds of manual modeling,instrument acquisition and image-based modeling.However,the image feature points of 3D face reconstruction based on single image are mostly visible.For the three-dimensional face under the influence of light,occlusion and attitude,Reconstruction technology still has modeling effects and efficiency that can not meet the needs of applications at the same time.This paper mainly solves the problem of three-dimensional face reconstruction through single image under self-occlusion,and realizes the end-to-end3 D face reconstruction algorithm.The main tasks include:1.An optimization algorithm based on self-masking single image feature point extraction is proposed,namely 3DDFA-ESDM algorithm.In order to avoid the ability of the parametric shape model to limit the deformation of the face,the HOG feature is extracted by using the convolution neural network to fit the input image into the 3D face model.The 3D face is used to realize large face masking.As an input,the linear regression ESDM algorithm is used to further improve the positioning accuracy of 2D feature points.Experiments show that the new algorithm has obvious effect on the feature points of the two-dimensional face images in self-occlusion,compared with the predecessors.2.In order to reconstruct the feature points on the basis of the extracted feature points,we also return the deformation of the 3D faces simultaneously,and propose a3 D face reconstruction algorithm based on the feature point regression.The algorithmis based on the strong correlation between the 2D feature point and the 3D face shape,so that the two-dimensional face image feature point positioning is coupled with the3 D face reconstruction process.Through the linear regression of two sets of cascades,To update the 2D feature point and the other group to update the 3D face shape.In each iteration,the ESDM algorithm is used to get the update of the feature point,and then update the 3D face shape according to the update of the feature point Volume,the whole process of 2D feature points,3D face shape updates are a rough to fine of the estimation process.Finally,the experimental results show that the validity of the three-dimensional face reconstructed by this algorithm solves the problem that the feature points generated by self-occlusion are not visible to affect the accuracy of 3D face reconstruction.