Evaluation of 3D mesh segmentation using a weighted version of the Ochiai index

Abstract

Despite several decades of research into 3D segmentation techniques and a diversity of methods and approaches proposed in the literature, comparing and evaluating the quality of this segmentation method is still a challenging task to achieve. In this paper, an objective evaluation metric suitable for evaluation of 3D segmentation quality and based on the Ochiai index, is proposed to enrich the stat of the art of this kind of method. In the proposed work, we adapt the Ochiai index to be able to compare an automatic segmentation with a ground truth one. The main added value of our method is the ability to treat regular and irregular meshes along with the ability to compare one automatic segmentation with many reference segmentations, these two functionalities, improve the performance of the proposed method and add more reliability to the obtained scores. Several experimental tests and a comparison with some commonly used methods have been made to show the potential and the performance of this approach.