Stratified helix information of medial-axis-points matching for 3D model retrieval

  • Authors:
  • Ji Jia;Zheng Qin;Jiang Lu

  • Affiliations:
  • Xi'an Jiaotong University, Xi'an, China;Xi'an Jiaotong University, Xi'an, China and Tsinghua University, Beijing, China;Xi'an Jiaotong University, Xi'an, China

  • Venue:
  • Proceedings of the international workshop on Workshop on multimedia information retrieval
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

With the increase of the number of available 3D models, the need for retrieving models from large databases to help people to find them has gained prominence. In this paper, we propose a method for automatic similarity estimation of 3D shape using stratified helix information of Medial-axis-points, which makes use of critical Medial-axis-points of skeletal instead of medial axes structure, and combines statistic and coordinate information of the Medial-axis-points. Skeleton is useful for many tasks including 3D shape retrieval, virtual navigation, reduced-model formulation, visualization improvement, mesh repair, animation, etc. But it has a high computational cost when applied in similarity estimation of 3D shape. The primary motivation for our approach are that the computational cost is lower than general skeleton method because it does not construct the whole skeleton and the shape matching issue is simplified to compare two vectors which may be different in length. The key idea is described as three steps. Firstly, the pose of triangular three-dimensional models is normalized and voxelized to volumetric objects. Secondly, we represent the signature of objects as stratified helix information of Medial-axis-points built by both Medial-axis-points based on repulsive force field function and the distribution of its rays. Lastly, we evaluate the similarity of objects by using multi-level weighted Hausdorff distance. The contribution of this paper is to utilize a stratified helix structure for every Medial-axis-point to establish stratified helix information, and introduce a variant Hausdorff distance for similarity measure. We find that our method is fast and easy to imply for discriminating between classes of 3D models in a test database. Also, we demonstrate that our method can provides useful discrimination of three-dimensional shapes, which is suitable as a pre-classifier for a recognition or similarity retrieval system.