Quantitative Comparison of Hole Filling Methods for 3D Object Search

Abstract

Retrieval of 3D models has become an important issue due to the increase in the number of digitized objects that are available in many different fields. When stored data present defects such as holes, accurate and reliable repairing tools are needed to solve these issues. In this work we present a comparative evaluation of hole filling algorithms from the local and global perspective, measuring quantitatively the quality of the repaired meshes and describing the impact these tools have on the models. We do this by mapping holes from one mesh onto another in order to create a synthetic dataset with realistic holes and ground truth and use the Hausdorff and RMS distance, as well as the mean angular deviation, to quantify the errors. The results show that the performance at a local level is similar for all compared methods, but large differences (up to 20%) appear when viewed at a global level, where algorithms that use volumetric representations introduce significant changes in the original models.