基于三维的人脸合成与替换系统实现  

Realization of 3D-Based Face Synthesis and Replacement System

作　　者：钟卓岑 史泽奇 

机构地区：[1]浙江理工大学信息学院,浙江 杭州

出　　处：《软件工程与应用》2022年第3期538-548,共11页Software Engineering and Applications

摘　　要：在保持人物身份的前提下对人脸合成,也称为人脸替换,目的将源图像中的人脸无缝且逼真地融合到目标图像。本文提出了一种自动化的人脸合成与替换系统,利用带有标签的图像合成用户感兴趣的全新形象,实现如虚拟试发、虚拟试衣的应用。系统的输入为单张人脸图像和查询文本,其中查询文本用于搜索相关图像,构建待替换的目标图像集合。为提高人脸合成效果,本文首先通过人脸图像重建三维人脸模型,在三维空间完成了人脸的姿态与表情迁移,然后通过人脸图像分解估计环境光照,实现了人脸图像的重光照,最后通过图像融合技术实现人脸的图像融合。实验证明我们的方法可以合成较为真实的人脸图像。Face synthesis, also known as face replacement, while maintaining the identity of the person, aims to seamlessly and realistically fuse the face in the source image to the target image. This paper proposes an automated face synthesis and replacement system, which uses labeled images to synthesize new images that users are interested in, and realizes applications such as virtual hair test and virtual fitting. The input of the system is a single face image and query text, where the query text is used to search for relevant images and construct a set of target images to be replaced. In order to improve the effect of face synthesis, this paper firstly reconstructs a 3D face model through face images, completes the face pose and expression migration in 3D space, and then estimates the ambient lighting through face image decomposition to realize the re-illumination of face images, and finally realize the image fusion of the face through the image fusion technology. Experiments show that our method can synthesize more realistic face images.

关 键 词：人脸替换 人脸合成 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]