Topology Matching for Fully Automatic Similarity Estimation of 3D Shapes

Masaki Hilaga, Yoshihisa Shinagawa, Taku Kohmura, Tosiyasu L. Kunii

Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques

Pages 203-212, 2001




Motivation

Recent technology has made it easy to construct complex 3D models of shapes. As a result, a large number of digital 3D models are now available, and there is a need to efficiently search through a large database of such shapes.

Goal of This Research

The authors wish to use a topological shape representation structure known as a Reeb graph to represent shapes. They then wish to search a database of 3D shapes by first comparing a query shape with each database shape using the Reeb graphs of the shapes.

Goal of This Paper

The authors introduce a technique called topology matching to measure the similarity of two shapes for the purposes of searching. The basic idea behind the technique is to compare two shapes by comparing their multiresolution Reeb graphs (MRGs), which is a set of Reeb graphs of a shape at different levels of detail.

Related Work

Results

The similarity estimation algorithm of this paper seems to be fairly accurate even if one of the shapes has been simplified (and the connectivity has hence changed), if noise has been added, or if one of the shapes has been transformed under a rigid motion. Searching based on this similarity estimation seems to yield results that are fairly consistent with human intuition. On average, the authors' search experiment took only 0.05 seconds to calculate one similarity, where one query shape was compared against 230 other shapes.

Bibliography



ctj