Skeleton Extraction of 3D Objects by Radial Basis Functions for Contentbased Retrieval in MPEG7

1
Skeleton Extraction of 3D Objects by Radial Basis
Functions for Content-based Retrieval in MPEG-7

• Ming Ouhyoung
• Fu-Che Wu, Wan-Chun Ma,
• Communication and Multimedia Lab
• Dept. of Computer Science and Information
  Engineering,
• National Taiwan University

2
Previous Collaboration MPEG-4 Based

• French Driven Talking Head and Lip Motion
  Analysis
• Realistic 3D Facial Animation Parameters from
  Mirror-Reflected Multi-view Video
• MPEG-4 Based
• Demo of the above ( ananova,anim2, track)

3
MPEG-4 to MPEG-7 Transition

• A New 3 Year Collaborative Project in Multimedia
  Lab, National Taiwan University
• Skeleton Extraction of 3D Objects for
  Content-based Retrieval in MPEG-7

4
(No Transcript)
5
3D Object Retrieval
Target model
Similarity
Search results

• 445 3D obejcts
• http//3dsite.dhs.org/dynamic

6
Part 1 3D Object Retrieval

• For each vertex of the object, calculate a sum of
  the geodesic distance from this vertex to others
• Get a Reeb graph, where each node represents a
  region according to the value
• The similarity of two objects is calculated using
  area and length of the node of their Reeb graph

7
Previous Work

• Masaki Hilaga, Yoshihisa Shinagawa, Taku Kohmura
  and Tosiyasu L. Kunii, Topology Matching for
  Fully Automatic Similarity Estimation of 3D
  Shapes, Proceedings of ACM SIGGRAPH, 2001.
• Robert Osada, Thomas Funkhouser, Bernard Chazelle
  and David Dobkin Matching 3D Models with Shape
  Distributions, Proceedings of Workshop on
  Shape-Based Retrieval and Analysis of 3D Models,
  Princeton, USA, Oct. 2001.

8
Previous Work

• Christopher M. Cyr and Benjamin B. Kimia, 3D
  Object Recognition Using Shape Similiarity-Based
  Aspect Graph, 2001.
• Michael Elad, Ayellet Tal and Sigal Ar, Content
  Based Retrieval of VRML Objects A Iterative and
  Interactive Approach, 2001

9
Using a skeletal structure of a 3D shape as a
search key

• Reeb graph
• Always consists of a one-dimensional graph
  structure
• Invariant to translation, rotation and scaling
• Robust against connectivity changes caused by
  simplification, subdivision and remesh
• Resistant against noise and certain changes due
  to deformation
• Introduce a multiresolutional structure

10
Geodesic distance

• The distance from point to point on a surface
  (the length of shortest path)
• Lazarus et al. proposed a level set diagram (LSD)
  structure in which geodesic distance from a
  source point is used as the function µ

11
3D Object Retrieval

• The approach represents the skeletal and
  topological structure of a 3D object
• Search 3D object automatically and quickly
• Robust against translation, rotation, scaling,
  simplification, subdivision, noise, deformation
• Demo
• http//3dsite.dhs.org/dynamic
• 445 objects in the database
• 0.08 sec for comparing two objects on the average

12
3D Object Retrieval
Target model
Similarity
Search results

• 445 3D obejcts
• http//3dsite.dhs.org/dynamic

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
Skeletal Representation by Radial Basis Function

• Improvements over Multi-resolution Reeb Graph
  not exactly a skeleton of mesh models
• How about Medial Axis Transformation
  Representation?
• Skip to part 2 slides