基于扩展总变差正则项的三维网格模型修复算法  

Mesh Inpainting Using Extended Total Variation Regularization

作　　者：翟羽佳 吴晓群[1,2] Zhai Yujia;Wu Xiaoqun(School of Computer and Information Engineering,Beijing Technology and Business University,Beijing 100048;Beijing Key Laboratory of Big Data Technology for Food Safety,Beijing 100048)

机构地区：[1]北京工商大学计算机与信息工程学院,北京100048 [2]食品安全大数据技术北京市重点实验室,北京100048

出　　处：《计算机辅助设计与图形学学报》2019年第11期1889-1897,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61602015);北京市教委科技一般项目(KM201910011012);“十三五”时期北京市属高校高水平教师队伍建设支持计划(CIT&TCD201904037)

摘　　要：针对特征保持的三维网格模型孔洞修复问题,提出一种基于扩展总变差正则项的修复算法.首先,根据邻接三角形中边界边的性质识别孔洞边界,利用动态规划方法重构孔洞区域的连接关系;然后,建立适用于三维网格模型修复的变分优化模型;最后,引入增广拉格朗日方法求解变分模型,迭代地优化三维网格模型的顶点位置.以带有孔洞的三维网格模型为数据,与2种基于体素的修复算法以及1种基于曲面的修复算法进行对比实验,实验结果表明,该算法能够有效地修复孔洞区域特征,在保持三维网格模型原始特征的同时全局地重建整个模型.A feature-preserving inpainting algorithm using extended total variation is presented to repair incomplete 3 D mesh models. Firstly, hole boundaries are detected based on the adjacent triangles and a dynamic programming method is used to build connectivities of the missing area. Secondly, a variational optimization model based on extended total variation is adapted to 3 D mesh models inpainting. Finally, an augmented Lagrangian method is employed to solve the introduced variational model to get the optimal vertex positions. The 3 D mesh models with holes are used as experimental data and comparison experiments are carried out on two volume-based algorithms and one surface-based algorithm. Experimental results demonstrate that our algorithm could recover holes well and reconstruct the whole models while preserving the initial features of the models.

关 键 词：三维网格模型修复 扩展总变差正则项 增广拉格朗日方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]