A Global Feature for Non-Rigid 3D Shape Retrieval with Shape Similarity

In the area of 3D shape retrieval, research about how to compare faster and more accurate is an important topic. 3D shape retrieval can be used in computer-aided design, video games, advertising and many other fields. This article proposes an enhanced 3D shape retrieval feature named 'parted Angle-Distance histogram (pAD) ', which is (1) robust over isometric deformations, (2) tolerant of topological and geometrical errors and degeneracies, and (3) able to be computed for non-manifold or 'polygon-soup' models, for searching surface-based 3D shape models. Pad is produced by three steps, (1) model conversion, (2) model clustering, and (3) 2D histogram generation. Two experiments for pAD are conducted: computing performance test and retrieval accuracy test, and they showed significant performance improved from preceding features.