Human Face Appearance Estimation With A Consumer RGB-D Camera

With the emergence and popularity of consumer RGB-D camera in recent years,ordinary users can easily obtain geometric information of various objects and people in real life.However,the objects in the real world are rich and colorful.Only the three-dimensional model of the geometry is obviously not enough to express the realism of the model.It is hard to reconstruct realistic model appearance with customer RGB-D camera,because it is general with low resolution,which makes its information not accurate enough.Besides,unknown material and lighting information are coupled in the color images which makes the problem more complex.In response to the increasing demand of realistic models,this paper aims to find a method to reconstruct the material of human face.We use the 3D model and a group of color images captured by customer RGB-D camera as input,and conduct a joint optimization for inaccurate camera pose,human face geometry as well as the unknown material and lighting information.Due to the complex structure of human skin,simulating the appearance of human skin is a very challenging issue.What's more,human beings are familiar with their faces,making it easy to draw any attention to any subtle artifacts of human face,which calls for more accurate modeling for human face material.This paper model the translucency of human skin with bidirectional surface scattering reflectance distribution unction(BSSRDF)and the specular reflection with physically based bidirectional reflection distribution function(BRDF).In order to obtain the initial value of face material more accurately,this paper also uses a priori knowledge of some face material parameters.Through experimental comparison and data analysis,the results show that the algorithm prevented in this paper can recover a more realistic face visual appearance.