Page 1
1
Modeling withSimplicial Diffeomorphisms
Luiz VelhoIMPA – Instituto de Matemática Pura e Aplicada
( PhD work of Vinicius Mello )© Luiz Velho SGP 2005 2
Outline• Background / Motivation• Framework Overview• Simplicial Diffeomorphisms• Binary Multi-Triangulations• Applications• Open Issues
© Luiz Velho SGP 2005 3
New Framework• Combines
– Parametric Representation– Implicit Surfaces
with– Spatial Warping
• Also…– Subdivision– Point Sets
© Luiz Velho SGP 2005 4
Background• Surface Descriptions
– Parametric– Implicit– Points
• Deformations– Structured– Unstructured
© Luiz Velho SGP 2005 5
Parametric Surfaces
• Piecewise Descriptions– Polygonal Meshes– Algebraic Patches– Subdivision Surfaces
• Other (Variational, etc…)
g :U ! R2" R
3
© Luiz Velho SGP 2005 6
Polygonal Meshes
+ Simple- Not Smooth
Page 2
2
© Luiz Velho SGP 2005 7
Parametric Patches
+ Smooth- Regular Mesh Topology
© Luiz Velho SGP 2005 8
Subdivision Surfaces
+ Arbitrary Base Mesh Topology- Extraordinary Vertices
© Luiz Velho SGP 2005 9
Implicit Surfaces
• Primitives• Compositions
– Combination of Algebraics– Volumetric
f!1
(c), f : R3" R
c f!1
© Luiz Velho SGP 2005 10
Volumetric Data
+ Local Control- Not Smooth
© Luiz Velho SGP 2005 11
Algebraic Implicits
+ Smooth- Restricted Control
© Luiz Velho SGP 2005 12
Point Sets
+ Simple- No Topology
(need local approximants)
Page 3
3
© Luiz Velho SGP 2005 13
Deformations
• Structured– Grid-Based
• Unstructured– Feature-Based
T :R3! R
3
T
© Luiz Velho SGP 2005 14
Grid-Based Deformations• Free-Form Deformations (parametric *)
• A-Patches (implicit)
Multiple Sheets
© Luiz Velho SGP 2005 15
Feature-Based Deformations• Wires (parametric)
• Clay (parametric)
© Luiz Velho SGP 2005 16
Pros & Cons
-- / ✓-++Algebraic
----/+-Volumetric
✓✓+-/++Subdivision
✓✓+-+Patches
✓✓++-Polygonal
FeatureGridcontroltopologysmooth
© Luiz Velho SGP 2005 17
Motivation“Combine Parametric and Implicit Surface
Descriptions with Spatial Deformations”
• Exploit:– Complementary Aspects– Integrated Representation
* Disclaimer:It does not solve all problems…
© Luiz Velho SGP 2005 18
Framework
G(u)=W ! X ! g(u)
F(p)= f ! X
!1!W
!1(p)
barycentric mapping
simplicialdiffeomorphism
implicit / parametric
X -1
X W
W -1
g
f
isocontour
Page 4
4
© Luiz Velho SGP 2005 19
Comparison• Deformation + Global Implicit Function
• Deformation of the Implicit Function
• Simplicial Deformation
F(X
!1(p))
F(w)
F!(X!
!1(W
!1(p)))
© Luiz Velho SGP 2005 20
Simplicial Diffeomorphism
• N-Dimensional Simplex :
• Standard N-Simplex :
* Properties of
– Maps Δ to σ, leaving faces invariant.
– Differentiable, has Differentiable Inverse
!= X(!
n)
!
!n
!
n= X
"1(!)
X
© Luiz Velho SGP 2005 21 © Luiz Velho SGP 2005 22
Why such Diffeomorphism is needed?
• Preserves Simplicial Structure– Maps faces to faces of a simplex
• Maintains the Topology of Level Surfaces– Continuous and Bijective
• Induces Smooth Geometry– Differentiable Map
© Luiz Velho SGP 2005 23
Curvilinear Iso-Simplicial Complex
1. Manifold M, triangulated by simplicial complex K
2. Function f from the vertices of K to R - {0}
3. Simplicial Diffeomorphisms for
X!:!
n" !
n
f1 f3
f6 f5
f2
f4
f(p) =0
!! K
© Luiz Velho SGP 2005 24
Simplicial Model• Implicit Simplicial Model
• Parametric Simplicial Model
F(p) = f (vi ) Xi,!
!1(W!
!1(p))
i=0
n
"
G(u) =! W
!(X!(g(uk )))
Page 5
5
© Luiz Velho SGP 2005 25
Main Issues- How to guarantee that X is a
Simplicial Diffeomorphism ?
- How to construct a suitableSpace Decomposition K ?
* Generality (dimension / degree)
© Luiz Velho SGP 2005 26
Conditions for Diffeomorphism
• A map is a diffeomorphismiff. for all
1.
2.
3.
X : K ! K
!! K
X(!)= !
X
int(! )is differentiable
D(X
int(! ))(p) is injective for all p! int(!)
© Luiz Velho SGP 2005 27
Simplicial-Invariant Function* Condition (1)
Definition:Let K be a simplicial complex anda continuous function.X is simplicial-invariant w.r.t. to K, or K-invariant,if for all
• Properties– Maps faces of K to themselves– Vertices remain fixed
X : K ! K
!! K , X(!) = !
© Luiz Velho SGP 2005 28
Differentiability and Injectivity* Combines conditions (2) and (3)
General Results: (Meisters-Olech)Let X be a differentiable function in .If X is injective in and locally injective inthen X is globally injective.
!
!" int(!)
© Luiz Velho SGP 2005 29
Characterization• Theorem:
A Simplicial Diffeomorphism X is:
1. A K-invariant function
2.
X : K ! K
det J
X> 0 in !! K
© Luiz Velho SGP 2005 30
Polynomial Simplicial Diffeomorphisms• Barycentric Representation
• H(w)=(H0(w), H1(w), …, Hn(w))where Hk is a Homogeneous Polynomial of degree m
1 1 … 1
p0
1p1
1! p
n
1
" " # "
p0
np1
n! p
n
n
!
"
###########
$
%
&&&&&&&&&&&&
H0(w)
H1(w)
"
Hn(w)
!
"
###########
$
%
&&&&&&&&&&&&&
=
1
X1(x)
"
Xn(x)
!
"
###########
$
%
&&&&&&&&&&&&
Page 6
6
© Luiz Velho SGP 2005 31
b101
Bernstein-Bézier Diffeomorphisms
• Control Points
• Bernstein-Bézier Polynomials
Hk
= bI
I =m
! BI
bI! A
n= {w! R
n+1| w
i=1}"
BI
=m
I
!
"###
$
%&&&&W
I
b110
b020
b002b200
b011
© Luiz Velho SGP 2005 32
Revisiting the Conditions B.B. Polynomials Conditions on
– Simple– Easy to Compute
• Looking Ahead…– Condition (1): Restrict Position– Conditions (2,3): Several Special Cases
! bI
© Luiz Velho SGP 2005 33
K-Invariance for B.B. Polynomials• Definition:
• Theorem (Adjusted Mapping):
* Intuition– Control Points bI are restricted to their faces
Let !(a0,…,a
m) = (e
0,…,e
m) where e
i= 1 if a
i> 0,
ei=!1 if a
i< 0 and e
i= 0 if a
i= 0
H is K-invariant if !(b
I) = !(I )
© Luiz Velho SGP 2005 34
Injectivity for B.B. Polynomials• Theorem (Pólya):
• Apply Pólya with* Particular Cases
– Degree 2 Polynomials– Central Control Point
Composition / Stratification
Let F ! R[W ] a homogeneous polynomial of degree m.
Then F(w) > 0 for all w!"n iff F #(W 0+…+W
n )M
has for some M !$ the form aKW
K
|K |=m+M
% with aK
> 0.
F = !H
© Luiz Velho SGP 2005 35
Quadratic Case (m = 2)• Theorem:
* No additional restrictions on bI
Let H = bIBI be a degree 2 adjusted polynomial.
|I |=2!Then H
"m is a "n -invariant diffeomorphism.
© Luiz Velho SGP 2005 36
Central Control PointDegree m = dim(Δ) +1
• Theorem:
* Restriction to move only the Central Control Point
Let H = bIBI be an adjusted degree m polynomial
|I |=m!and J = (J 0 ,…, J n )"#m+1 with J = m and J i " {0,1}.
If bI
= I / m for I $ J, then H%m is a %n -invariant diffeo.
Obs: Nice Handle
Page 7
7
© Luiz Velho SGP 2005 37
Composition* The Composition of Simplicial Diffeomorphisms
is also a Simplicial Diffeomorphism
• Definition:
Obs: Raises degree Composition is not commutative in general
Let F = {F!}!!K
and G = {G!}!!K
be simplicial diffeomorphisms.
The composition FG = {F!G!}!!K
of F and G is obtained by
applying F !G in each simplex !! K .
© Luiz Velho SGP 2005 38
Stratified Scheme• Composition G =G1…Gn of n mappings Gj which
correspond to actions in each topological dimension
* Preserves Symmetry of bI
Proposition:
Gj
i=Wi + (bJ
i! J
i/ J
J = j+1,J k "{0,1}
# )BJ
G1, G2 and G3 are !3-invariant diffeomorphisms.
© Luiz Velho SGP 2005 39
Wrap-Up• Summary
– Powerful Scheme– Flexible and General– Arbitrary Dimension
• Examples– 3D
© Luiz Velho SGP 2005 40
© Luiz Velho SGP 2005 41
Space Decompositions* Need Decomposition of Ambient Space
to Construct Iso-Simplicial Complex
• Desired Properties– Semi-Regular Tiling– Adaptation Power– Gradual Transitions between Cells– Natural Multi-Resolution Structure
© Luiz Velho SGP 2005 42
Stellar Operators• Basic Transformations on a Simplicial Complex
Flip
Split / Weld
cell
edge
Page 8
8
© Luiz Velho SGP 2005 43
Some Results of Stellar Theory• Theorem (Newman, 1931):
Two n-dimensional simplicial complexes are piecewisehomeomorphic iff they are related by a finite sequenceof elementary stellar operations.
• Proposition:Any stellar operation can be decomposed by a sequenceof stellar operations on edges.
© Luiz Velho SGP 2005 44
Binary Multi-Triangulations• Definition:
A binary multi-triangulation is a poset (T, >),where T={M0, M1,…, Mk} is a finite set of simplicialn-complexes and the order relation > satisfies:
1. There is a maximum, and a minimum in T;
2. if and only if, for a stellar operation ξ on some edge A
* Best Properties (Puppo & De Floriani, 1998)
M M
!M !M M
! (A)! "!! #M
© Luiz Velho SGP 2005 45
Graph Representation of BMT• BMT is a Lattice
• Nodes are sub-meshes, arrows are stellar operations
• Any separating cut in the graph is a valid complex© Luiz Velho SGP 2005 46
Adaptive Simplicial Decomposition* Every n-simplex, , has a refining element
(split edge or flip edge) and an unrefining element(weld vertex or flip edge).
• Algorithm: adapt(e) for all ti incident in e do if element(ti) != e then adapt(element(ti)) apply(stellar, e)
Enforces Transition (Restricted Structure) Same Algorithm for Local Refining and Unrefining
t ! M "M
© Luiz Velho SGP 2005 47
Construction Methods Base Complex
– Semi-Regular Structure
• Subdivision– Refinement
• Simplification– Coarsening
• Dynamic Adaptation– Refinement and Coarsening
© Luiz Velho SGP 2005 48
Applications
• Free-Form Modeling
• Surface Approximation
• Reconstruction from Points
Page 9
9
© Luiz Velho SGP 2005 49
Free-Form Modeling
© Luiz Velho SGP 2005 50
Free-Form Modeling• Current Implementation
– One Control Point per k-Face– Iterate Mappings Two Times
• Future Plans– Hermite Handles (Normal + Point)– Natural Continuity Constraints (same Normal)
© Luiz Velho SGP 2005 51
Surface Approximation 2D
© Luiz Velho SGP 2005 52
Surface Approximation 3D
© Luiz Velho SGP 2005 53
Surface Approximation• Fitting Based on Implicit Function
• Optimization Method– L-BFGS-B(Zhu, Byrd and Nocedal)
min f (X!(bI )
(pi ))!!T
"2
s.t.
bI !#I $1
pi
© Luiz Velho SGP 2005 54
Reconstruction from Points* Under Development
• Similar to Implicit Surface Fitting– Optimization Based on Distance to Samples
• Need to Compute Distance from Surface– Project Points onto Current Approximation
Page 10
10
© Luiz Velho SGP 2005 55
Open Issues• Continuity
• Rendering
• Polynomial Approximation
• Rational Diffeomorphisms
© Luiz Velho SGP 2005 56
Thanks!