Research On Automatic Human Face 3D Transplanting

It is always an important subject of computer vision to capture,analysis and process human faces.With the wide spreading of digital cameras and smart mobilephones,the capture of 2D human faces is getting within reach.The capture of 3D faces is relatively difficult.The major route to capture a 3D face is via a 3D scanner.3D scanners based on white light or lasers are harmful to eyes and skin inspite of high resolution.The effective distance of depth cameras based on infra-red are getting more and more closer in recent years.Intel company has launched a depth camera called the RealSense,with the effective distance of less than 30 centi-meters.So with the help of the algorithms such as KinectFusion,it is feasible to scan faces via a depth camera.The face data captured is usally be applied in the areas of encoding,monitoring or entertainment,besides the fundamental function of recording information.Recently,with the fast development of social media,a lot of face transplanting applications have been raised.The 2D case of face transplanting applications include face swapping,photograph composition,privacy protection and so on.The 3D case applications are comparatively more practical.It can be applied to generate facial patches for large 3D games.It can also be used for 3D printing or some other real scenarios.The existing 3D face transplanting applications are rare,and can be with the problems of poor similarity and excessive confine to the given database.The capture of a 3D face includes the collection of not only the color information,but also the 3D geometry information,and the matching relationship between them.It is also important to consider these aspects to guarantee the results of 3D face transplanting.The basic indicators of 3D face transplanting are similarity,seamless fusion,features matching and less distortion.This paper presents an automatic 3D face scanning and transplanting system.This system is composed of scanning module and face transplant-ing module,and these two modules are seamlessly connected.The major contribution of the scanning module is to use the depth camera to scan a 3D face model with texture quickly,and to crop it according to the feature points to acquire a relatively standard one.The main contribution of the face transplanting module is that it is without any human interactions,and with well compatibility.In the aspect of color,this system uses a seamless image blending algorithm based on image gradient to pre-process the tones of the images,and then it uses an optimization method to deal with the chromatic aberration.A mesh Laplace preserved method is raised to deform the 3D face,and then it stitches with the background model automatically.The advantages of this system are it can achieve realistic effects and it can be applied to any 3D characters with textures.