Face Reconstruction Based On 3D Morphable Model

In recent years,the data volume processed by computer vision is increasing with the rapid development of computer technology,and 3D models are widely used in various fields.With the popularization of face unlocking technology in mobile phone,the technology of obtaining 2D face information has been relatively mature.The 3D face reconstruction has become a research hotspot in the field of computer graphics.At present,many researchers use 2D face image features combined with 3D Morphable model to improve 3D reconstruction method.Most of them are reconstruction method by single two-dimensional image.Once the lack of information from multiple perspectives,facial reconstruction effect often has the problem of generalization,but the morphable model calculation based on multiple views always has big data quantity,and long time-consuming.Based on the 3D Morphable model,the 3D face reconstruction algorithm in this paper improves the key steps such as data acquisition,model fitting,texture mapping and reconstruct the 3D face reconstruction process with multiple views.The details are as follows:The primary task of 3D face reconstruction is to obtain 2D face information,including 2D landmarks and texture data.In order to make the reconstructed model more precise,it is necessary to obtain landmarks accurately.In this paper,CLNF algorithm with three-layer convolutional network is adopted to identify 2D face landmarks.Dense alignment model determines the initial shape of face model.We use Surrey Face Model with expression parameters to enrich the data of 3D Morphable Model.In the stage of model fitting,twice linear regression is designed to solve blendshapes coefficient and shape coefficient to achieve the fitting morphable from 2D landmarks of face and 3D statistical model.For texture mapping,we made light treatment to the model firstly,and then extract the gather texture information with ISOMAP algorithm,marked the visibility parameters of each pixel point,and realized the omni-directional texture mapping by combining the current facial attitude,try to avoid the phenomenon of "drawing" and "cavity" in 3D reconstruction model.For the images collected by the same person from different angles,we designed a 2D landmarks to track the change of face angle,and a new linear iterative fitting method to achieve all-round accurate reconstruction.Finally,the algorithm efficiency and reconstruction effect of this paper are compared with the current reconstruction methods of the same type.The experimental results show that our method has better reconstruction effect in the face reconstruction process,and the accuracy and time efficiency are significantly improved.