1
Surface Parameterization A Tutorial and Survey Michael Floater and - - PowerPoint PPT Presentation
Surface Parameterization A Tutorial and Survey Michael Floater and - - PowerPoint PPT Presentation
Surface Parameterization A Tutorial and Survey Michael Floater and Kai Hormann Presented by Afra Zomorodian CS 468 10/19/5 1 Problem 1-1 mapping from domain to surface Original application: Texture mapping Images have a
Afra Zomorodian
2
Problem
- 1-1 mapping from domain to surface
- Original application:
Texture mapping
– Images have a natural parameterization – Goal: map onto surfaces
- Geometry processing
– Approximation – Remeshing – Data fitting
- Input: Piecewise (PL) triangular meshes
Afra Zomorodian
3
History
- Ptolemy (100-168)
- Preserve:
– angles – area
- Loxodrome: constant bearing
Orthographic Mercator (1512-1594) Stereographic Lambert (1728-1777)
Afra Zomorodian
4
Theory
- Gauß (1777-1855)
– differential calculus – differential geometry
- Riemann (1826-1866)
- Riemann Mapping Theorem:
– Input: any simply-connected region of complex plane – Output: any other simply-connected region of complex plane – Statement: there exists a map that preserves angle
Afra Zomorodian
5
Maps
- Parameterized surface S 3
- Regular:
– xi are smooth (C) – partials are linearly independent
- First fundamental form
Afra Zomorodian
6
What is I?
- I is:
– symmetric (3 DOF) – g = det I = g11 g22 – g12
2 > 0
– positive definite
Afra Zomorodian
7
Types of Mappings
- Isometric:
– preserves lengths – developable surfaces
- Conformal:
– preserves angles – Stereographic and Mercator projections
- Equiareal:
– preserves area – Lambert projection
- (Theorem) isometric conformal + equiareal
scalar
Afra Zomorodian
8
Planar Mappings
- f : 2 2
- f(x, y) = (u(x,y), v(x,y))
- I = JTJ, J is Jacobian of f
- Singular values of J σ1, σ2 are square roots of
eigenvalues λ1, λ2 of I
Afra Zomorodian
9
The Game
- Try to find
– isometric maps (only developable surfaces) – conformal maps: no distortion in angles – equiareal maps: no distortion in area
- Impossible?
– try to minimize distortion (maybe a mixture) – define some sort of energy function (sometimes implicit) – minimum is your answer – easy to compute
Afra Zomorodian
10
Outline
- Continuous
– Conformal – Harmonic (Distortion minimizing) – Equiareal
- Discrete
× Not:
× Mean Value Coordinates (6.3) × Boundary mapping (6.4) × Linear methods (7.3) × Closed surfaces (9)
Afra Zomorodian
11
- View as Conformal map function of complex variable
ω = f(z)
- Locally, analytic in neighborhood of z, f (z) ≠ 0
- z = x + iy, w = u + iv
- Conformal maps satisfy
Cauchy-Riemann equations:
- Laplace equations:
- Laplace operator:
- Harmonic Maps satisfy Laplace equations
- isometric conformal harmonic
A Complex View
Afra Zomorodian
12
Harmonic Maps
- RKC Theorem:
– f harmonic, S* 2 (convex) – f maps ∂ S homeomorphically to ∂ S* f is 1-1
- Just map boundary!
- Approximate PDEs
- Inverse not harmonic
- harmonic conformal
Afra Zomorodian
13
Minimizing Distortion
- No guarantee on angles
- Harmonic maps minimize Dirichlet Energy:
- For surface S 3, Generalized Laplace
- Laplace-Beltrami operator
Afra Zomorodian
14
Equiareal Maps
- Conformal maps are almost unique
- Example: map unit disk onto self
– Choose z S, angle φ – By Riemann mapping theorem, unique f : S S, f(z) = 0 and arg f’(z) = φ – 3 degrees of freedom: complex number z (2) and φ (1)
- Lots and lots of equiareal maps (badly behaved, too)
scalar
Afra Zomorodian
15
Outline
Continuous
- Discrete
– Harmonic – Conformal – Equiareal
Afra Zomorodian
16
Discrete Surfaces
- Surface S 3
- PL surface ST = {T1, …, TM}
- Polygonal domain S* 2
- PL map f : ST S*, f linear on each Ti
- Uniquely determined by image of vertices
Afra Zomorodian
17
Discrete Harmonic Maps
- Finite Element Method
– fix boundary somehow – minimize Dirichlet energy for internal vertices
- Quadratic minimization
- (Nice) Linear System
Afra Zomorodian
18
- Discrete harmonic maps tend to harmonic maps
- 1-1: non-degenerate triangles that are not flipped
- Normalized weights
- Linear system
- If wij positive, so are λij
- Convex Combination Maps
- (Theorems) Discrete harmonic maps are 1-1:
– Barycentric Maps (1/di) [Tutte, 3-connected graphs] – Any weights such that j Ni λij = 1 – If opposite angles sum < π (eg Delaunay)
Convex Combination Maps
Afra Zomorodian
19
- Unlike harmonic, ∂ does not to be fixed
- Condition number of Jacobean:
- Discretizing:
- Minimum is 2 number of triangles
- For PL: conformal isometric (developable)
- Most Isometric Parametrizations (MIPS)
Discrete Conformal Maps
Afra Zomorodian
20
- Relationship between EM and ED
- Non-linear minimization
- MIPS energy
– on degenerate triangles – on ∂(Ki) (star-shaped neighborhood of vi)
- Local functional is convex, so use Newton’s method
Computing MIPS
Afra Zomorodian
21
Angle-Based Flattening
- φi: angles in ST
- αi: angles in S*
- φ(v): sum of φi around vertex v
- For interior vertices, α(v) = 2π
- Optimal angles βi = φis(v)
- Minimize
- Non-linear constraints
Afra Zomorodian
22
Discrete Equiareal Maps
- Minimize energy functional?
- Badly behaved
- Multiple minima with E(f) = 0
- Minimize mixture of energies
Afra Zomorodian
23
Take Home
- Isometric: only developable
- Conformal: preserve angles
- Harmonic: minimize Dirichlet energy
- Equiareal: preserve area, but promiscuous
- Angles mean we deal with tangents (partials)
- Area means we deal with Jacobian
- In practice, find right energy to minimize
- Theory is good!