mesh resampling

46
Mesh Resampling Wolfgang Knoll, Reinhard Russ, Cornelia Hasil 1 Institute of Computer Graphics and Algorithms Vienna University of Technology

Upload: tass

Post on 24-Feb-2016

55 views

Category:

Documents


0 download

DESCRIPTION

Mesh Resampling. Wolfgang Knoll, Reinhard Russ, Cornelia Hasil. 1 Institute of Computer Graphics and Algorithms Vienna University of Technology. Motivation. Reducing number of faces while trying to keep overall shape , volume and boundaries Oversampled 3D scan data - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Mesh Resampling

Mesh Resampling

Wolfgang Knoll, Reinhard Russ, Cornelia Hasil

1 Institute of Computer Graphicsand Algorithms

Vienna University of Technology

Page 2: Mesh Resampling

Reducing number of faces while trying to keep overall shape, volume and boundaries

Oversampled 3D scan data

Fitting isosurfaces out of volume datasets

Motivation

2Cornelia Hasil

Page 3: Mesh Resampling

Motivation

Simplification useful tomake storagetransmissioncomputationdisplay more efficient

Can reduce memory requirementsand can speed networktransmission

3Cornelia Hasil

Page 4: Mesh Resampling

Problem Statement

Transform a given polygonal mesh into another with fewer faces, edges, and vertices:

4Cornelia Hasil

Page 5: Mesh Resampling

Mesh Simplification Approaches

Two basic concepts

Vertex Clustering

Incremental Decimation

Example

Incremental decimation with quadric error metric

5Cornelia Hasil

Page 6: Mesh Resampling

Mesh Simplification Approaches

Vertex ClusteringCluster GenerationComputing a representative Fast and effective Poor quality

Uniform 3D gridMap vertices to cluster cells

Remove degenerate triangular cells

Computing a representative:If P1, P2, ..., Pk are vertices in the same cell, then the representative is P = (P1 + P2 + ... + Pk)/k

6Cornelia Hasil

Page 7: Mesh Resampling

Mesh Simplification Approaches

Incremental decimationGeneral

Repeat pick mesh region apply decimation operator

Until no further reduction possible

7Cornelia Hasil

Page 8: Mesh Resampling

Example: Quadric Error Metrics

Surface Simplification Using Quadric Error Metrics

Iterative Pair Contraction with the Quadric Error Metric

Works on non-manifold geometry

Supports aggregation

Can be implemented efficiently

Produces high quality approximations

8Cornelia Hasil

Page 9: Mesh Resampling

Surface Simplification Using Quadric Error Metrics

Pair contraction: (v1 , v2 ) → v ̄

A pair of vertices (v1, v2) are valid for contraction if: 1. (v1, v2) is an edge, or2. ||v1 − v2|| < t for some threshold t

BenefitsCan join unconnected components Can result in much nicer approximations

9Cornelia Hasil

Page 10: Mesh Resampling

Approximating Error With Quadrics

For each vertex vi store a symmetric 4x4 matrix Qi

Error (v) at v = [vx vy vz 1]T is vTQvv The matrices Qi are called quadrics, because the level sets of (v) = ε form quadric surfaces (usually ellipsoids)

For a given contraction(v1 , v2 ) → v ̄ , let Q ̄ = Q1 + Q2

10Cornelia Hasil

Δ

Δ

Page 11: Mesh Resampling

Performing Contractions

To perform a contraction(v1 , v2 ) → v ̄ , we must find v ̄

Simple scheme: select v1, v2 or (v1 + v2)/2 with lowest value for (v ̄ )

We find minimum v ̄ by solving ∂ /∂x = ∂ /∂y = ∂ /∂z = 0which is equivalent to

11Cornelia Hasil

Δ

Δ

Δ

Δ

Page 12: Mesh Resampling

Algorithm Summary

Compute initial quadrics for each vertex

Select all valid pairs

Compute optimal contraction target for each pair and let its associated error be the cost of the contraction

Place all pairs in a keyed heap

Iteratively remove the pair with least cost from the heap, contract the pair, and update the cost of all valid pairs involving this contracted vertex

12

Page 13: Mesh Resampling

Error Metrics in Mesh Simplification

Reinhard Russ

Institute of Computer Graphics and Algorithms

Vienna University of Technology

Page 14: Mesh Resampling

The purpose of using Error Metrics

Measurement for the introduced geometric error

What is the best contraction to perform?

What is the best position for the remaining vertices?EndpointOptimization

14Reinhard Russ

Page 15: Mesh Resampling

Wolfgang Knoll 15

Metrics Simple Heuristics

Edge length, Dihedral angle, area etc.

Sample Distance Squared distance function Cluster distance function

Curvature Valence function Quadratic Error Metric Feature Sensitive Metric

Page 16: Mesh Resampling

Simple Heuristics as Error Metrics

Edge length

Edge marking function

Dihedral angleSurrounding area

16Reinhard Russ

Page 17: Mesh Resampling

Sample distance as Error Metrics

Squared distance function

Cluster distance function

17Reinhard Russ

Page 18: Mesh Resampling

Cluster distance function

18Reinhard Russ

Page 19: Mesh Resampling

Sample distance as Error Metrics

Projection to closest point

Restricted projection

19Reinhard Russ

Page 20: Mesh Resampling

Error Metric based on Curvature

Curvature

20Reinhard Russ

Page 21: Mesh Resampling

Error Metric based on Curvature

Curvature Tensor FieldHow many lines should be traced on the surface?

Compute local densitySpacing between two lines of curvatureCross section of the surface (normal to the lines)

21Reinhard Russ

Page 22: Mesh Resampling

Error Metric based on Valence function

Valence function

22Reinhard Russ

Page 23: Mesh Resampling

Quadric Error Metric

Quadric Error MetricBased on point-to-plane distance (instead of point-to-point distance)Minimize sum of squared distance to all planes at vertex

23Reinhard Russ

Page 24: Mesh Resampling

Quadric Error Metric

Quadric Error MetricConstruct a quadric Q for every vertexCompute error of collapsingCompute quadric for new vertex

24Reinhard Russ

Page 25: Mesh Resampling

Quadric Error Metric Adaptations

Originally for ECP-based algorithmsAdaptations for VDP-based and FCP-based algorithms

Originally for Polygonal ModelsAdaptations for Point Clouds

25Reinhard Russ

Page 26: Mesh Resampling

Feature Sensitive Metric

Consider the field of unit normal vectors as a vector-valued image

26Reinhard Russ

Page 27: Mesh Resampling

Metric Classification

Wolfgang Knoll

Institute of Computer Graphics and Algorithms

Vienna University of Technology

Page 28: Mesh Resampling

Wolfgang Knoll 28

Groupings Goal

Simplification/Minimization Quality Improvement Topology/Feature Preservation

Application Area Approach

Page 29: Mesh Resampling

Wolfgang Knoll 29

Simplification/Minimization Less vertices, triangles, faces etc. than before

→ Smaller Mesh

Either: Given amount Minimal under error boundary

Combinable with other goals

Page 30: Mesh Resampling

Wolfgang Knoll 30

Simplification/Minimization

Use of metrics depending on the methods.

Page 31: Mesh Resampling

Wolfgang Knoll 31

Simplification/Minimization

Use of metrics depending on the methods.

But:Almost every metric can be used for Minimization...

Page 32: Mesh Resampling

Wolfgang Knoll 32

Simplification/Minimization Edge/Vertex/Region decimation based on simple

Heuristics Clustering in regards with Energy minimization Etc...

Page 33: Mesh Resampling

Wolfgang Knoll 33

Quality Improvement Improvement in regards with:

Vertex/Face distribution Connectivity Triangle shape

Keeping Quality in regards with: Error-Metric Topology/Feature Preservation

Page 34: Mesh Resampling

Wolfgang Knoll 34

Quality Improvement Not always quantizable!

Metrics: Curvature based metrics QEM, squared distance function and other

metrics for minimizing error/energy(taking the best choice)

Valence function

Page 35: Mesh Resampling

Wolfgang Knoll 35

Quality Improvement

Example: Redistribution

Page 36: Mesh Resampling

Wolfgang Knoll 36

Topology/Feature Preservation Not necessary error minimization

Goal is to keep topology intact and/or maintain important features

More triangles at feature-areas

Rules for Topology preservation

Used metrics are often curvature-based

Page 37: Mesh Resampling

Wolfgang Knoll 37

Topology/Feature Preservation

Example: Feature preservation

Page 38: Mesh Resampling

Wolfgang Knoll 38

Application Area Dependent on:

Operational area Local Global

Geometrical Element Vertex Edge Face

Page 39: Mesh Resampling

Wolfgang Knoll 39

Application Area Local approaches are often based on decimation

approaches with simple heuristics

Global approaches: (Iterative) Energy minimization Clustering

→ both often use distance metrics Feature Sensitive Metric

Page 40: Mesh Resampling

Wolfgang Knoll 40

Application Area Distance metrics (obviously) use mostly the distance

between vertices/points

Clustering additionally can be extended with the Curvature around vertices and in an area/face

Page 41: Mesh Resampling

Wolfgang Knoll 41

Application Area

Example: global method

Page 42: Mesh Resampling

Wolfgang Knoll 42

Approach Decimation Approaches

Energy minimization

Clustering

Other Approaches:Particle Simulation, Retiling etc.

Page 43: Mesh Resampling

Wolfgang Knoll 43

Comparison Measures Used to analyze and compare a method

Often tightly tied with the method goal

Usually no direct comparison between proposed methods due to different measures & models

Page 44: Mesh Resampling

Wolfgang Knoll 44

Comparison Measures Quantitative Goal:

Size: vertex/face number, # Bytes Speed: computational performance

Qualitative Goal: Error: max, avg Triangle-angle

Page 45: Mesh Resampling

Wolfgang Knoll 45

Comparison Measures Quality improvement often lowers performance

Size reduction often lowers quality

Page 46: Mesh Resampling

Wolfgang Knoll 46

Conclusion Main goal is minimization!

→ reducing the numbers

Clustering and Decimation approaches

Curvature often used for quality improvement

Performance often goal dependent