一种基于NL-SFS方法的三维人脸快速重建系统

针对快速重建三维人脸的需求,设计了一种基于非Lambert从明暗恢复形状(NL-SFS)方法的三维人脸快速重建系统。首先人脸表面采用非Lambert反射模型描述其反射特性,这更接近于人脸表面实际的反射特性;接着由遵循正交投影的摄像机获取光源作用下的人脸表面图像,同时假定摄像机方向与光源方向保持一致,建立人脸表面的图像辐照度方程;然后将该方程转化为包含人脸表面深度信息的Eikonal方程;最后根据上述方程的特点,利用fast marching方法设计了系统软件,能够快速求得Eikonal方程的解,进而重建出人脸的三维形状。实验结果表明,该系统可以在较短的时间内获得较高的重建精度,即0.9s内可达到0.43%的高度平均相对误差。

A fast recovery system of 3D face based on NL-SFS approach

Due to the demand of fast recovery of 3D face, a recovery system based on non-Lambertian shape from shading(NL-SFS) approach was designed. Firstly, the non-Lambertian reflectance model was used to characterize the face, which was very closed to the actual reflectance property. Then, under a light source the face image was acquired by a camera whose projection mode was orthographic. The image irradiance equation for the face was derived under the assumption that the direction of the camera was the same as the light source. Furthermore, the equation was formulated into an Eikonal equation which included the shape of the face. Lastly, according to its characteristic, the system software was developed by using the fast marching method. The software can compute the solution of the equation, and then the 3D shape of the face was obtained. Experimental results demonstrated that the designed system could fastly achieve high recovery accuracy, which is 0.43% mean relative error within 0.9 s.