A New Evaluation Method for Mesh Segmentation Based on the Levenshtein Distance

Abstract

3D mesh segmentation is an important preprocessing in many 3D shape applications and it is considered as one of the most challenging tasks in 3D mesh analysis. Recently, interests have been developed in creating different approaches for 3D mesh segmentation and many solutions have been proposed in the literature. Hence, one of the most important issues is how to evaluate the proposed segmentation methods and judge the quality of segmentation results. In this paper, we present a new evaluation metric, based on the Levenshtein distance to perform an objective evaluation of 3D mesh segmentation methods. Most of the existing metrics judge the quality of an automatic segmentation with only one ground-truth, which is considered as a limit since we can have many semantic segmentations for the 3D meshes. Furthermore, they do not take into consideration the regularity of meshes. Therefore, the novel metric allows both comparison with multiple ground truth segmentations and has the ability to give a relevant evaluation for both regular and irregular 3D meshes. Several experiments have been done to show the performance and the potential of the proposed metric.
Copyright © 2016 Praise Worthy Prize - All rights reserved.