Linear Regression Facial Expression Reconstruction And The Application In Facial Animation

The research of facial expression animation loved by researchers is still a hot research subject in the field of multimedia technology.The purpose of the research about facial expression animation is to use the human facial pose in the animation and facial muscle movements to exposit a character's expression and emotion.Real-time and authenticity are currently two important indicators of monitoring facial expression animation.In the production process of traditional facial expression animation,the realistic image of the animation often takes enormous human effort,not being able to achieve real-time facial expression.So it becomes an urgent requirement to find a more efficient expression animation technology,and the performer's facial expression animation is born.The synthesis technology on performer's facial expression animation is that let models perform the expression that we needed.By using camera,sensor,Kinect or a pattern of software processing to record the dynamic of human facial expression,we can effectively extract the expression parameter by using parameter data to drive animation models and create expression animation.This can reduce the time of the boring work by men and the production cost to the expression animation,completely guaranteeing the facial-expression's authenticity.Facial expression animation synthesis technology driven by performing mainly includes two key steps:facial expression motion capture and facial expression animation compounding.In terms of facial expression motion capture,this paper used Delaunay triangulation to triangulate for facial expression model by Kinect capture.Not only can Kinect work in the situation of not having enough light,but it also has invariance on texture and color,which can provide more deep information.But for Kinect the captured facial expression pattern is consisted of amount of point cloud,messy and covering densely.Sometimes,part of the data lost,the data cannot be directly used for the later face alignment and the establishment of expression based.So paper used Delaunay triangulation to triangulate,which makes the face point cloud into a triangular mesh.In facial animation synthesis,this paper proposed a automatic generation method of face alignment and individual facial expression based on local and global feature regression.The method used the local feature to estimate its shape and added the global feature as the constrain condition.In the cascade mode,the face alignment is trained to make the estimated face shape as same as the real one.The generation of personalized facial expressions is essential for facial animation.The facial expression model collected by Kinect is used to generating a set of initial expression based on a set of a priori model.And then refine them step by step.Next construct a set of personalized expression sets.Finally,the facial expression animation is formed by the deformation transfer,which is transferred to the model to become facial expression animation.Based on two improvements,facial expression animation has achieved good performance on two different benchmark datasets.