一种用于三维人脸重建的快速稠密视差图生成方法  

A Method of Fast Generating Dense Disparity Map for 3D Facial Reconstruction

作　　者：谷天[1] 陈江龙[2] 李科[2] 曾东[2] 

机构地区：[1]四川大学华西口腔医学院,四川成都610041 [2]电子科技大学生命科学与技术学院,四川成都610054

出　　处：《西华大学学报（自然科学版）》2016年第1期77-79,98,共4页Journal of Xihua University:Natural Science Edition

基　　金：四川省科技支撑计划项目(2013GZ0117;12ZC0220);2012年四川大学创新训练计划(20120193)

摘　　要：口腔数字化是口腔临床发展的重要方向,三维颅颌面部结构数字化重建是其重点环节。为此,文章提出一种用于三维人脸重建的稠密视差快速求解的方法。该方法对校准后的图像进行初始匹配,构建强对应点点集,再对点集内的点进行三角剖分,并分配初始视差,逐步求精生成稠密视差。该方法加入强匹配对应点,减少了匹配二义性的产生,同时有效克服了因使用全局能量函数优化方法而易陷于局部最小值的缺点。实验结果显示,该方法匹配速度快、精度高,能满足三维人脸重建的实际应用需求。The craniofacial and oral digitalization is an important area in the dental clinical practice. And the method of craniofacial and oral reconstruction is very important. This paper focuses on the stereo matching of 3D face reconstruction, and a triangulationbased fast dense disparity map generation method is proposed according to the smooth and low texture features. Firstly, we matched images calibrated and obtained a set of strong corresponding points . Then we triangulated the points in the point set in order to build a linear system and gradually minimized the energy function in the polar direction and refined to yield a dense disparity by assigning the initial disparity. This method can reduce the match ambiguity through adding strong matched corresponding points, and at the same time avoid the disadvantage of easily getting trapped in local minimum due to the global energy optimization. The result shows that this method is fast and highly precise and can meet the need of practical application.

关 键 词：人脸重建 立体视觉 稠密视差 三角剖分 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]