基于图像的脸部模型调整及表情动画技术研究  

Image-based 3D Model Adaptation for Face Reconstruction and Expression Animation

作　　者：朱云峰[1] 刘跃虎[1] 苏远歧[1] 郑南宁[1] 

机构地区：[1]西安交通大学人工智能与机器人研究所,陕西西安710049

出　　处：《系统仿真学报》2006年第z1期409-413,共5页Journal of System Simulation

基　　金：国家自然科学基金项目(60021302;90412010);863高技术项目(2005AA147060)

摘　　要：采用一种基于图像的三维人脸模型主动调整方法,实现了三维特定人脸模型的真实感重建和表情动画生成。通过特定人脸的正面和正侧面图像上的93个人脸表情特征点,采用整体调整及基于径向基函数插值的局部调整方法对通用三维人脸模型进行几何调整,并利用变形拼接及多分辨率图像融合技术生成无缝的人脸纹理镶嵌图,实现具有真实感的三维特定人脸模型重建。改进了人脸肌肉物理模型并结合弹性约束模型,避免了在表情动画处理中由于肌肉收缩力过大导致网格三角片毁坏而破坏人脸的拓扑结构问题,其局部网格动态细分策略,使生成的表情动画更加细腻。An image-based model adaptation method was adapted for realistic facial model reconstruction and expression animation. Firstly, 93 feature points were extracted from both front and side images of an individualized face. Secondly, global geometrical adaptation was performed according to the correspondence of the feature points and the generic 3D facial model, and then, 3D generic facial model was transformed with the Radial Basis interpolation to fit the specific facial image as a local adaptation. After those, texture mapping was used to reconstruct 3D specific facial model, and texture image was generated by multi-resolution mosaic. Subsequently, the muscular model and elastic repulsion model were employed for generating facial expression animation and retaining facial topological. Dynamic local subdividing on mesh triangles was used for enhancing reality.

关 键 词：三维人脸建模 径向基插值 多分辨率图像融合 肌肉模型 弹性约束模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]