基于梯度光图像的高精度三维人脸重建算法  

High Precision Three-Dimensional Face Reconstruction Algorithm Based on Gradient Light Image

作　　者：黄硕 胡勇[1,2] 巩彩兰[1,2] 卜弘毅[1,4] 顾耕[4] 许琦[4] 王薇[4] Huang Shuo;Hu Yong;Gong Cailan;Bo Hongyi;Gu Geng;Xu Qi;Wang Wei(Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China;Key Laboratory of Infrared System Detection and Imaging Technology,Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China;Jiaxing Opto-Electronic Engineering Center,Chinese Academy of Sciences,Jiaxing,Zhejiang 314022,China)

机构地区：[1]中国科学院上海技术物理研究所,上海200083 [2]中国科学院上海技术物理研究所红外探测与成像技术重点实验室,上海200083 [3]中国科学院大学,北京100049 [4]中国科学院嘉兴光电工程中心,浙江嘉兴314022

出　　处：《光学学报》2020年第4期38-46,共9页Acta Optica Sinica

基　　金：上海市科委项目(17411952800,18441904500)。

摘　　要：针对人的脸部特征复杂、细节纹理丰富,获取高精度人脸三维模型成本高、工作量大的问题,提出基于梯度光图像的高精度三维人脸重建算法。利用球形灯光装置建立6种不同的光照模式,并利用偏振光特性对镜面反射光和漫反射光进行分离;利用单向梯度光图像,并基于双向反射分布函数(BRDF)得到反射光方向与法线方向的关系,通过计算获取镜面反射和漫反射法线图;将增强后的镜面反射和漫反射法线图进行融合,并映射到低精度模型上,可以重建出高精度的三维人脸模型。实验结果表明,本文方法计算复杂度低,重建精度高,窗口大小为7×7的高通滤波增强的镜面反射图和漫反射法线图以2…1的比例融合之后,重建的效果最佳,模型能精细到微尺度级。To address the problems of complex and rich details of face features and heavy work and high cost of reconstructing high-precision three-dimensional(3 D)face models,a high-precision 3 Dface reconstruction algorithm based on the gradient light image is proposed.Six different illumination modes are established using a spherical light device,and specular and diffuse reflection lights are separated using the polarization characteristic.Using a unidirectional gradient light image,specular and diffuse normal maps are obtained by calculation based on relationship between the reflection and normal directions by employing the bidirectional reflectance distribution function(BRDF).The obtained maps are merged and mapped to the low-precision mode to reconstruct a highprecision 3 Dface model.The experimental results show that the proposed method has low computational complexity and high reconstruction accuracy.Based on the obtained results,it is best to use the high-pass filtering with a window size of 7×7 enhanced specular and diffuse reflection normal maps to achieve the best reconstruction at 2…1 map ratio fusion.The proposed model can be refined to microscale.

关 键 词：图像处理 三维重建 梯度光 双向反射分布函数 法线贴图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]