基于形变模型的三维人脸重建方法及其改进  

3D Face Reconstruction Based on the Improved Morphable Model

作　　者：胡永利[1] 尹宝才[1] 谷春亮[1] 程世铨[1] 

机构地区：[1]北京工业大学多媒体与智能软件技术实验室,北京100022

出　　处：《计算机学报》2005年第10期1671-1679,共9页Chinese Journal of Computers

基　　金：国家自然基金(60375007);北京市自然科学基金(D070601-01)资助.~~

摘　　要：形变模型(morphablemodel)是近几年出现的三维人脸建模新方法.该方法使用原型人脸的组合表示新的人脸,对于特定人脸图像,通过模型匹配实现了三维人脸的自动重建.虽然形变模型具有自动化、真实感好等优点,但现有形变模型的建立依赖于不稳定的人脸图像对应光流算法,模型匹配只考虑了一般光照环境下的人脸重建问题,且建模计算量大.针对以上问题,文章对形变模型进行了改进:提出了网格重采样的方法,实现了模型人脸数据的精确对应;建立了多分辨率的三维人脸模型;在模型匹配过程中采用了多光源光照模型,使模型可适用于复杂光照环境下的人脸重建.实验结果表明,上述改进可以有效提高模型匹配的效率和准确性以及模型对光照的适应性.In the past years, a novel face modeling method, the morphable model is proposed to reconstruct realistic 3D face from images. The model uses the combination of prototypic 3D faces to represent new faces. Matching the model to the given facial image, the 3D face of the image will be reconstructed automatically. Although the model has good result in face modeling, there are some disadvantages. The construction of the model depends on the unstable algorithm, the optical flow algorithm, which does the pixel-wise correspondences among the prototypes. The model is not fit for the complex illuminations and requires too much computation in the model matching. This paper proposes a mesh resampling method to do the correspondences and construct a multi-resolution model to reduce the computation. A multi-light lighting model is devised to overcome the model matching problems in complex illumination. The experimental results show above methods have impressive effectiveness on the improvements of the model matching accuracy, efficiency, and illumination adaptability.

关 键 词：形变模型 三维人脸重建 网格重采样 光照模型 光流算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]