基于网格变形的从图像重建三维人脸

从图像重建高质量三维人脸一直是计算机视觉和图形学的一个重要研究问题.不同于传统的基于立体匹配的窄基线多视几何和数据驱动的人脸形变方法,提出一种结合网格变形技术和立体视觉原理的、从图像重建高质量三维人脸模型方法.给定从不同视角拍摄的几幅人脸图像,基于健壮图像特征获得可靠的相机外部参数和稀疏三维点;在此基础上,提出一种结合几何细节保持和图像一致性约束的三维人脸变形算法重建三维人脸,通过对人脸模板的网格变形,使得变形人脸在多幅图像中的可见投影具有一致性的图像颜色强度.基于模板的人脸变形可以有效地解决三维模型成像中的遮挡问题,采用健壮估计法消除噪声、离群点和光照对目标函数收敛性的影响,对目标函数的多次非线性优化求解进一步改进了人脸重建的质量.采用合成人脸图像和真实人脸图像重建三维人脸的实验结果表明,文中算法可以从几幅宽基线图像重建高质量的三维人脸模型.

Mesh Deformation Based 3D Facial Modeling From Images

High quality 3D facial modeling from images is an important problem in computer vision and graphics.Unlike traditional narrow-baseline multi-view geometry and data driven shape blending facial modeling from images,we propose a new technique for high quality facial modeling from images by deforming a template mesh model using Laplacian mesh deformation and color consistency metric defined by all input images.Given a few facial images taken from different viewpoints,we first do sparse stereo matching from matched robust image feature points to recover camera extrinsic parameters and a few 3D matched points.Then an image based mesh deforming process is employed to deform a 3D facial template to make the color intensities of visible images consistent.Template based deformation avoids the difficulty due to the occlusion in images.In addition,the robust estimation technique reduces the influence of image noise,outliers and surface highlights.Multi-stage nonlinear optimization improves the quality of facial modeling.Experimental results using synthetic and real images show that our method provides stable and quality facial reconstruction results from wide-baseline images.