改进Phong光照模型的人脸去高光与重光照算法  

Face Highlights Removal and Relighting Algorithm Based on Improved Phong Illumination Model

作　　者：黄颖[1] 王泽荃 李焱晖 黄江平 HUANG Ying;WANG Zequan;LI Yanhui;HUANG Jiangping(School of Software Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065)

机构地区：[1]重庆邮电大学软件工程学院,重庆400065

出　　处：《计算机与数字工程》2024年第12期3715-3721,共7页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61572092)资助。

摘　　要：针对目前人脸重光照算法无法准确反映人脸反射情况的问题,提出了一种关于改进的Phong光照模型的人脸图像去高光算法与重光照算法。该算法主要分为三部分:1)依据该光照模型,可渲染得到合成数据集用以训练神经网络去除高光。2)利用半监督学习将去高光能力转移到真实人脸中。3)将去高光后人脸图像进一步分解为光照相关和无关分量。替换光照相关部分并重组图像,实现人脸重光照。实验结果表明,改进后的算法可提升重光照精确度与视觉效果。Aiming at the problem that the current face relighting algorithm can not accurately reflect the face reflection,a face image specular removal algorithm and a relighting algorithm based on the improved Phong illumination model are proposed.The algo-rithm includes three parts.1)According to the illumination model,rendered synthetic data to train the neural network to remove highlights.2)The semi-supervised learning is used to transfer the ability of specular removal to real faces.3)After specular removal,face image is decomposed into components related to illumination and component unrelated to illumination.The component related is replaced to illumination ande the image is recombined to achieve face illumination.Experimental results show that the improved algorithm can improve the relighting accuracy and visual effect.

关 键 词：基于图像的重光照 深度学习 光照模型 高光去除 逆渲染 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]