controlled-distortion constrained global parametrization
DESCRIPTION
Controlled-Distortion Constrained Global Parametrization. Ashish Myles Denis Zorin New York University. Global parametrization. Map surface to plane Quadrangulation , remeshing Geometry images Smooth surface fitting Texture/bump/displacement mapping. Goal: singularity placement. - PowerPoint PPT PresentationTRANSCRIPT
1
Controlled-Distortion Constrained Global Parametrization
Ashish MylesDenis Zorin
New York University
2
Global parametrizationMap surface to plane
Quadrangulation, remeshing Geometry images Smooth surface fitting Texture/bump/displacement
mapping
3
Goal: singularity placement
singularities(valence ≠ 4)
4
Seamless global parametrizationRequirementsFeature alignment Map feature curves
to axis-aligned curves
Seamlessness All curvatures
multiples of π/2
Parametrize Quadrangulate
cone singularities(quad valence 3)
π/2 π/2
Integrated curvature(angle deficit) π/2
Seam-constraints
Feature linegrid-aligned
0 curvature (flat)
5
Cones for features and distortion
Distortion induced by feature-alignment
Cones required toalign to features
π/2
0 – π/2
π/2
6
Distortion vs # conesProposition 5. For a closed smooth surface with bounded total curvature, and any ϵ > 0, there is a seamless parametrization which has total metric distortion EARAP < ϵ.
7
Goal: automatic cone placementConstraints: (1) seamless (2) feature-alignedAutomatically determine cone locations + curvatures (mixed integer problem)
Trade off parametric distortion with number of cones.
No conesHigh distortion
Bad
2 conesLow distortion
Good
8
Cross-field smoothing
Easy feature alignmentNo direct distortion control
Conformal maps
Easy to measure distortionOver-constrained for multiple features
Contribution: Unification of techniques
scale by eϕ
cross-field rotation ∗ω = dϕ + ∗h closest conformal map
ϕ2
9
Smooth cross-field not distortion-aware
10
Smooth cross-field not distortion-aware
parametricdomain collapsed region
To avoid collapse
23 cones
31 cones
11
Seamless + feature-alignment constraints1. Internal curvature
–∇2ϕ = Knew – Korig
2. Feature line curvature ∇nϕ = κnew – κorig
3. Relative alignment constraints
Conformal maps cannot handle condition 3.But cross-fields can.
Conformal maps: over-constrained
π/2
0
π/2 boundary curvature
0 boundarycurvature
Surface alreadysatisfies 1 and 2.
Completelydetermine ϕ
π/2
– π/2
12
Relation of conformal to cross-fields
Simple 2D case∇θ⊥ = ∇ϕ
(∇θ computes field rotation)
Conformal map
scale by eϕ
rotate by eiθ
3D surfaces (exterior calculus) ∗ω = dϕ
(Connection form ω ∼ ∇θ) [Crane et al. 2010]
13
Conformal + Cross-field
Interpretation of harmonic part: Not seamless! Distortion ||ϕ||2 only
accurate if ||h||2 low
Need to minimize ||h||2
cross-fieldrotation
closestconformal map
harmonic part
Constant scale across
∗ω = dϕ + ∗h
14
System setup and solveCompute ∗ω = dϕ + ∗h (determines cross-field)
Minimize energy ||ϕ||2 + α ||ω||21. Satisfy linear constraints for seamlessness and alignment2. Minimize ||h||2 (gives linear equations)
Mixed Integer problem Integer variables (curvature, relative-alignment) Solve via iterative rounding (similar to [Bommes 2009])
Linearconstraints
Quadratic energy(cross-field smoothness)(distortion)
15
α = 0.00190 cones
α = 0.0166 cones
α = 0.147 cones
α = 1.044 cones
α = 10032 cones
no harmonic minimization
Field smoothing[Bommes 2009]
32 cones
0.0
0.6
16
Comparison to cross-field smoothing
Field smoothing[Bommes 2009]
Ours 0.0
0.6
50 cones
72 cones
17
Comparison to cross-field smoothing
0.0
0.6Field smoothing[Bommes 2009]
Ours
18
Comparison to cross-field smoothing
0.0
0.6
19
Field smoothing[Bommes 2009]
Ours
Comparison to cross-field smoothing
To avoid collapse
0.0
0.623 cones
31 cones
20
Comparison to cross-field smoothing
0.0
0.8
30 cones
42 cones
Field smoothing[Bommes 2009]
Ours
21
SummaryUnification via ∗ω = dϕ + ∗h of
1) cross-field smoothing and2) conformal approach
Support both1) feature alignment and2) distortion control
Trade-off between1) distortion and2) # cones
22
Questions?