摘要
针对传统方法存在的通用性不强、实时性较差和与视频传输系统不兼容等缺点,提出一种面向单幅任意姿态图像的3D人脸建模技术.首先将基于初始位置校正和模型实例选择的主动外观模型改进方法应用到人脸特征点的提取过程中;然后结合人脸结构特征和空间仿射变换调节CANDIDE-3线框模型,实现对应人脸的全局位置恢复和形状匹配.在此基础上,根据感兴趣区域的对称性对局部坐标进行微调,并构建真实感纹理.实验结果表明,该技术不受图像拍摄焦距等因素影响,单幅图像平均建模时间约为300ms.
A 3D face modeling approach for a single facial image with any pose is proposed in this paper. This paper aims to overcome the shortcomings of traditional methods such as weak universality, lack of real time, non-compatible with video transmission systems, and so on. Firstly, an improved active appearance model algorithm based on the initial position adjusting and model instance selection is applied to extract facial feature points, then we adjust candied-3 wire-frame model to match global location and shape using structural characteristics of actual face and spatial affine transformation. Finally, we tune the local coordinates according to the symmetry of region of interest and construct realistic texture. Experiments results show that the proposed approach is not sensitive to focal length and the average time consumption of single image is about 300 ms.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2009年第5期668-673,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60672099)
教育部留学回国人员科研启动基金
西南交通大学2009博士研究生创新基金
关键词
主动外观模型
位置恢复
形状匹配
线框模型
3D人脸建模
active appearance models
location restoration
shape match
wireframe model
3D face modeling
