An objective 3D mesh segmentation evaluation using Sokal-Sneath metric

It is widely recognized that the segmentation of 3D objects is one of the most important disciplines in computer vision. In the last few years, developments of 3D segmentation techniques are in continual expansion. However, little work has been directed toward the evaluation of 3D mesh segmentation methods and consequently there is still no satisfactory performance measure. In this paper, a new evaluation method proposed to evaluate 3D segmentation algorithms. The proposed method is based on Sokal-Sneath metric to determine the degree of dissimilarity between an automatic segmentation and a set of references. This new method offers two important advantages, it allows a comparison between an automatic segmentation and a set of ground truth segmentations instead of only one reference and it can be applied on regular meshes as well as irregular meshes. Different experiments are presented in this paper to validate the performance of the proposed approach.