一种基于反射特性的人脸三维重建方法

Recovery of 3D Shape Based on Reflectance Characteristics

不同姿势下的人脸识别研究中的一种重要思路就是在三维空间解决姿势变化问题,如何由二维人脸图像得到三维人脸形状是其中的一项关键技术。本文提出了一种由两幅正面人脸图像估计人脸表面反射特性,重建三维形状的算法。首先根据人脸的对称特性,由两幅给定光照的正面人脸图像,估计得到个性化人脸特征:反射特性(反射系数与法向量的乘积),并提出消除鼻梁阴影的改进算法;然后由反射特性重建三维人脸形状,最后根据估计得到的反射特性和三维形状合成不同姿势下的人脸图像,实验结果表明,本文提出算法合成的人脸图像更自然一些,而且合成图像不受训练集图像的影响。本方法计算简单,并且不需要任何三维人脸数据作为先验知识。

one of approaches to face recognition with pose variation is to deal with pose variation in 3D space, but how could we get the 3D shape of a face from corresponding 2D face images is a key issue. In this paper, a new approach is proposed to estimate the reflectance characteristics and 3D shape from two frontal face images. Based on the symmetry of face surface, the subject-specific infor- mation： reflectance characteristics （combining the albedo and normal vector） are estimated from two frontal face images with given illu- mination direction, furthermore the method is improved to eliminate the pseudo-shadow near nose-bridge in the generated face image by threholding. Then the 3D shape is recovered by reflectance characteristics. Finally, the face images with specified pose are synthesized, the experimental results shows that the synthesized face images from different training images are nature and almost same, especially, the computation is simple and 3D face data are unnecessary in proposed approach.