Face Editing And Bas-Relief Modeling Based On Face Alignment

With the development of society,face detection and face alignment are hot topics in computer vision,and they are also the basis of three-dimensional face reconstruction,virtual makeup,expression analysis and other related applications.This paper has conducted the following research based on face alignment technology:(1)skin color detection is the process of identifying the human skin region pixel image in an image or video.However,due to skin color detection affected by light,color,and other environmental conditions,the area of skin positioning has been a difficult issue.A new skin detection technique combining a Gaussian mixture model and a face alignment technique is proposed.Firstly,the face skin area is roughly located,then the adaptive thresholding algorithm,morphological algorithm and connected domain algorithm are used to accurately detect the facial skin.In this paper,by defining 8 pairs of facial features,the clustering algorithm combining SOM and K-Means is used to classify the face shapes,and then the color transferring algorithm based on Lab color space is used to edit the skin colors.The face skin was edited,and finally the face value of the face was scored using a deep neural network based on ResNet50.(2)A method based on 2D face poses is proposed to drive the 3D model deformation.Firstly,the facial alignment technique is used to detect the facial features of the face in the 2D image.The iterative closest point fitting algorithm is used to obtain the rotation matrix between the image feature points and the 3D model feature points.Then,based on a skeleton-driven mesh deformation method,the face pose of a 2D image is used to drive the pose transformation of the 3D face model.Finally,a normal image of the model under the computed posture is calculated to generate a bas-relief model.