Errorbounded Reduction of Triangle Meshes with Multivariate Data

1
Error-bounded Reduction of Triangle Meshes with
Multivariate Data

• Chandrajit L. Bajaj
• Daniel R. Schikore

2
Outline

• Problem Definition
• Related Work
• Motivation
• Outline of Algorithm
• Results
• Future Work

3
Problem Definition

• Geometric surface defined by triangles
• Multivariate data defined at vertices
• Compute a reduced size mesh for display which
  adapts to both the geometry and the data defined

4
Related Work

• Geometric Reduction - Point/Edge/Triangle
  Deletion (Hamann, Rossignac et al., Schroeder
  et al.) Energy Optimization (Hoppe et al.,
  Turk) Resampling (He et al.) Remeshing (Eck et
  al.)
• Data Reduction - Point Insertion/Deletion
  (DeFloriani et al., Fowler et al., Lee, Puppo
  et. al, Silva et al., Tsai, Ware et
  al.)

5
Motivation

• Data is often defined on surfaces
• Existing mesh reduction methods satisfy error
  criteria in geometry or data, but not both.

6
Algorithm Outline

• Consider each vertex v for deletion
• Compute a triangulation of the neighbors of
  vertex v
• Determine mapping between triangulations
• Iteratively attempt to improve triangulation
  (reduce error) by edge-flipping
• Delete v if error criteria are satisfied

7
Mapping Between Triangulations

• Projection of New
• Triangulation

Interior Vertex
8
Error Representation

• Two error values per face - Easy to update
  - Combines information about all deleted
  vertices

9
Computing Geometric Error

• Geometric error is quantified by the signed
  distance from the old triangulation to the new
  triangulation
• Displacement in the direction of normal is
  positive error by convention

10
Computing Data Errors

• Errors in data are quantified by the signed
  difference between interpolated data values in
  the old and new triangulations

11
Error Propagation

• Projection of New
• Triangulation

Segmentation
12
Retriangulation

• Initial triangulation is arbitrary
• Cost of an edge may be computed as a) Maximum
  error in geometry or data b) Maximum error
  introduced in geometry or data
• Edges are flipped if cost of the flipped edge
  is lower than the cost of the current edge

13
Results
14
Projectile Impact
Original Data (2869 tri)
3 Error (1356 tri)
Data Courtesy Lawrence Livermore National
Laboratory
15
Projectile Impact
Original Data (2869 tri)
3 Error (1356 tri)
Data Courtesy Lawrence Livermore National
Laboratory
16
Earth Science Data
Original Data (4846 tri)
3 Error (2384 tri)
Data Courtesty Space Science and Engineering
Center
17
Earth Science Data
Original Data (4846 tri)
3 Error (2384 tri)
Data Courtesty Space Science and Engineering
Center
18
Future Work

• Improved measures of error (Linear/ Quadratic
  fitting)
• Better initial triangulation
• Improved edge-flipping for optimization