三维网格分割的错分评价准则

提出一种新的、基于面片错分率和面积错分率的三维网格模型分割定量评价准则.定量评价是精确衡量分割效果、针对特定应用选择最有效的分割算法、以及指导新算法研究的重要基础.基于分割质量显著的数据库进行的三维分割评价准则给出的定量评价指标属于模糊的、统计性质的评价,在评价特定类型的分割时,该评价指标的可信较低、精确性较差.本文基于普林斯顿大学数据库中7类385份高质量的手工分割结果,以及7种自动分割算法中分割数目与手工分割数目相近的部分高质量数据,基于面片错分和面积错分两种准则,对7种自动分割算法进行定量了评价.实验证明本文提出的两种错分评价准则具有较高的精度和可信性.

Misclassified Evaluation Metrics for 3D Mesh Segmentation

The quantitative evaluation metrics for 3D mesh segmentation has been more important in many applications of digital geometry processing,to select more suitable algorithm,and to guide the research of novel method.In this paper,we have proposed two novel criteria to quantitative evaluate 3D mesh segmentation: Face Misclassified Error and Area Misclassified Error.Due to the qualitative problem of consistency of human-generated segmentation,the quantitative evaluations provided by Princeton benchmark is fuzzy and statistic,the index of 7 segmentation algorithms is suspect and imprecise.Based on a subset of with 385 manually generated segmentations for 7 different object categories,and with the number of sub-mesh is very nearly the same,we quantitative evaluated 7 automatic segmentation algorithm of how well they perform using FME and AME.The experiment proved our metrics have more precision and creditability.