SAND2016-9194C

Photos placed in horizontal position

with even amount of white space
between photos and header

An Introduction to Automatic
Mesh Generation Algorithms
Short Course, September 26, 2016
Washington, DC
Steven Owen
Sandia National Laboratories
Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin
Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000. SAND NO. 2011-XXXXP

1
Agenda
Part I 8:00-9:30AM Part II 10:00-11:30AM
 The Simulation Process  Tet vs. Hex Meshing
 Geometry Basics  Structured vs. Unstructured
 Mesh Representations  Structured Hex Methods
 Mesh Generation Methods  Unstructured Hex Methods
 Tet/Tri Meshing Methods  Hex Dual Representations
 Surface Meshing Basics  Overlay Grid
 Smoothing  Automatic Block
Decomposition
 Hybrid Methods

2
Classical References
J. F. Thompson, B. K. Soni, and N. P. Weatherill, eds.,
Handbook of Grid Generation, CRC Press, 1998.
Block-structured grids:
• Transfinite-interpolation, Elliptic and hyperbolic PDE
systems, Harmonic mappings, ...
• Unstructured grids: Quadtrees and Octrees methods,
• Advancing-front methods, Delaunay-Voronoi methods,
• Anisotropic Grid generation, ...

P. J. Frey and P.-L. George, Mesh Generation - Application to

Finite Elements, Hermes Science Publishing, Oxford, UK, 1st
ed., 2000, 2nd ed. 2008.
• A comprehensive survey of Tetrahedral mesh generation
methods: Quadtree-octree methods, Advancing-front methods,
Delaunay-based methods, ...

3
Simulation Process

3 2 kN

1. Build CAD Model 2. Mesh 3. Apply Loads and

Boundary Conditions

4. Computational Analysis
5. Visualization

4
 Adaptive Simulation Process
User
supplies meshing
parameters
2
3 2 kN

1. Build CAD Model 2. Mesh 3. Apply Loads and Boundary Conditions

Adaptivity Loop

Error? Analysis Code

Error >  supplies meshing
parameters

6. Remesh/Refine/Improve
4. Computational Analysis 5. Error Estimation

Error < 

7. Visualization

5
Geometry

vertices: x,y,z
location

6
Geometry

curves: bounded by
vertices: x,y,z two vertices
location

7
Geometry

curves: bounded by surfaces: closed

vertices: x,y,z two vertices set of curves
location

8
Geometry

curves: bounded by surfaces: closed

vertices: x,y,z two vertices set of curves
location

volumes: closed set

of surfaces
9
 Geometry

Manifold Geometry:
Each volume maintains
Surface 11
its own set of unique
Volume 1
surfaces
Volume 2
Volume 2

Surface 7
Surface 8 Surface 9 Surface 10 Surface 11

Volume 1

Surface 1 Surface 2 Surface 3 Surface 4 Surface 5 Surface 6 Surface 7

10
Geometry

Non-Manifold
Geometry: Volumes
share matching surfaces
Volume 1

Volume 2
Volume 2

Surface 7
Surface 8 Surface 9 Surface 10

Volume 1

Surface 1 Surface 2 Surface 3 Surface 4 Surface 5 Surface 6 Surface 7

11
Geometry

Mesh Generation Algorithm

Geometry Engine

12
Geometry

Triangle Hex
Delaunay
Paving Advancing Sweeping
Tet Mesher
Front Tool

Geometry Engine

13
 Geometry

Triangle Hex
Delaunay
Geometry Query Paving Advancing Sweeping
Front
Tet Mesher
Tool
Topology Query
Curve::project_to() Surface::curves()
Surface::project_to() Curve::vertices()
Surface::normal_at() Volume::surfaces()
Surface::curvature_at() Vertex::curves()

Geometry Engine

14
 Geometry

Triangle Hex
Delaunay
Geometry Query Paving Advancing Sweeping
Front
Tet Mesher
Tool
Topology Query
Curve::project_to() Surface::curves()
Surface::project_to() Curve::vertices()
Surface::normal_at() Volume::surfaces()
Surface::curvature_at() Vertex::curves()

ACIS, OpenCascade, Catia, Facets, …

15
 Geometry

Triangle Hex
Delaunay
Geometry Query Paving Advancing Sweeping
Front
Tet Mesher
Tool
Topology Query
Curve::project_to() Surface::curves()
Surface::project_to() Curve::vertices()
Surface::normal_at() Volume::surfaces()
Surface::curvature_at() Vertex::curves()

CAD Abstraction
Open
ACIS Catia Facets
Cascade

16
Mesh Representation

mass quadrilateral
beam
triangle

0D 1D 2D

tetrahedra
wedge
hexahedra pyramid

3D

17
Mesh Representation
Node-Element Virtual Edges Cyclic Connectivity Doubley-Linked
Connectivity Only and/or Faces Connectivity

Node Node Node Node

Edge Edge Edge

Face Face Face

Element Element Element Element

Slower Speed Faster

Low memory Memory Usage High memory

18
Mesh Representation
Geometry Associativity

Vertex Node
Boundary Conditions applied to
geometric entities.

Delete/Modify mesh defined on

geometric entity
Curve Edge List
Operations on mesh entities
require conformity to geometry

Surface Face List

Volume Element List

19
 Mesh Generation Methods
Nodes
mesh projected to
surfaces geometry

mesh
vertices
curves

Geometry First Mesh First

20
Mesh Generation Methods

Geometry First Mesh First

21
Mesh Generation Methods
• AKA. Bottom-up Meshing • AKA. Overlay Grid
• Advancing Front • Octree Methods
• Delaunay Methods • Grid-based Methods
• Boundary constrained • No Boundary constraints
• Dimension n mesh is input for dimension • Boundary nodes placed based on
n+1 intersection and/or projection
• Incremental Workflow • All-in-one Workflow
• One entity at a time • All geometry at once
• Neighbors conform • Neighbors will not conform
• More control over mesh quality and • Less control over mesh quality and
element placement element placement
• Boundary layers • Orientation of initial mesh w.r.t. geometry
• Directionality effects final mesh
• Structured meshing requires special • No special case topologies for hex
case topologies meshing
• Pave-sweep • No special case decompositions needed
• Block-structured
• Manual/Automated decomposition required

Geometry First Mesh First

22
Mesh Generation Process

Apply Manual Verify/correct for

Mesh Mesh
Sizing, Match sizing criteria on
Vertices Curves
Intervals curves

For each surface

Set up sizing
Mesh Smooth/Cleanup Verify
function for
surface surface mesh Quality
surface

For each volume

Set up sizing
Mesh Smooth/Cleanup Verify
function for
volume volume mesh Quality
volume

Geometry-First Mesh Generation

23
Mesh Generation Process
Initialize Modify mesh to
overlay meet local size
mesh requirements

Match Match Match Match

nodes to edges to tris/quads tets/hexes
vertices curves to surfaces to volumes

Remove
Smoothing/
exterior
Cleanup
mesh

Mesh-First Mesh Generation

24
 Tri/Tet
Methods
Octree
Advancing Front
Delaunay http://www.simulog.fr/mesh/gener2.htm

http://www.ansys.com

25
Tet Meshing Software
A Representative Sample

Commercial
• Tetmesh-GHS3D, INRIA, Rocquencourt, Distene France.
• MeshSim, SCOREC, RPI, Simmetrix Inc. USA.
• SolidMesh, AFLR mesh generator, SimCenter, Mississippi State Uni.,
(Altair Hypermesh)

Open Source
• Netgen, TU Vienna.
• Gmsh, Uni. Liege & Uni. Catholique de Louvain.
• GRUMMP, University of British Columbia.
• Pyramid, UC Berkeley.
• CGALmesh, INRIA, Sophia-Antipolis.
• TetGen, Weierstrass Institute, Berlin.

26
Octree/Quadtree

•Define intial bounding box (root of quadtree)

•Recursively break into 4 leaves per root to resolve geometry
•Find intersections of leaves with geometry boundary
•Mesh each leaf using corners, side nodes and intersections with geometry
•Delete Outside
•(Yerry and Shephard, 84), (Shepherd and Georges, 91)

27
Octree/Quadtree

QMG,
Cornell University

28
Octree/Quadtree

QMG,
Cornell University

29
Advancing Front

A B
•Begin with boundary mesh - define as initial front
•For each edge (face) on front, locate ideal node C based on front AB

30
Advancing Front

C D
r

A B

•Determine if any other nodes on current front are within search

radius r of ideal location C (Choose D instead of C)

31
Advancing Front

•Book-Keeping: New front edges added and deleted from front as

triangles are formed
•Continue until no front edges remain on front

32
Advancing Front

•Book-Keeping: New front edges added and deleted from front as

triangles are formed
•Continue until no front edges remain on front

33
Advancing Front

•Book-Keeping: New front edges added and deleted from front as

triangles are formed
•Continue until no front edges remain on front

34
Advancing Front

•Book-Keeping: New front edges added and deleted from front as

triangles are formed
•Continue until no front edges remain on front

35
Advancing Front

r
C

A
B
•Where multiple choices are available, use best quality (closest
shape to equilateral)
•Reject any that would intersect existing front
•Reject any inverted triangles (|AB X AC| > 0)
•(Lohner,88;96)(Lo,91)

36
Advancing Front

37
Advancing Front

38
Advancing Front

39
Advancing Front

r
D

40
Advancing Front

r
D

41
Advancing Front

B D

42
Advancing Front

C
D

43
Advancing Front

Ansys, Inc.
www.ansys.com

44
Delaunay

Triangle
Jonathon Shewchuk
http://www-2.cs.cmu.edu/~quake/triangle.html
Tetmesh-GHS3D
INRIA, France
http://www.simulog.fr/tetmesh/

45
Delaunay

circumcircle

Empty Circle Property:

No other vertex is contained within the circumcircle of
any triangle
46
 Delaunay

Computing the Circumcircle center and radius

Ax  By  C  x 2  y 2  x1 , y1 
 x3 , y3 
 x1 y1 1  A  x12  y12 
x   
y2 1  B    x22  y22 
  xc , yc 
 2  r
 x3 y3 1 C   x32  y32 

A B
xc  , yc  , r  C  xc2  yc2
2 2  x2 , y 2 

47
47
Delaunay

Delaunay Triangulation:
Obeys empty-circle (sphere) property

48
Delaunay

Delaunay Triangulation:
Obeys empty-circle (sphere) property

49
Delaunay

Delaunay Triangulation:
Obeys empty-circle (sphere) property

50
 Delaunay

Vornoi cell

Delaunay Triangulation:
Obeys empty-circle (sphere) property

51
Delaunay

Voronoi Diagram
Dual of a Delaunay Triangulation

52
 Delaunay

circumsphere

Empty Sphere Property:

No other vertex is contained within the circumsphere of
any tetrahedron
53
 Delaunay

circumsphere

Empty Sphere Property:

No other vertex is contained within the circumsphere of
any tetrahedron
54
 Delaunay

Given a Delaunay
Triangulation of n nodes, How
do I insert node n+1 ?
X

Lawson Algorithm
•Locate triangle containing X
•Subdivide triangle
•Recursively check adjoining
triangles to ensure empty-circle
property. Swap diagonal if
needed
•(Lawson,77)

55
 Delaunay

Lawson Algorithm
•Locate triangle containing X
•Subdivide triangle
•Recursively check adjoining
triangles to ensure empty-circle
property. Swap diagonal if
needed
•(Lawson,77)

56
 Delaunay

Bowyer-Watson Algorithm
•Locate triangle that contains
the point
•Search for all triangles whose
X circumcircle contain the point
d (d<r)
r •Delete the triangles (creating
c a void in the mesh)
•Form new triangles from the
new point and the void
boundary
•(Watson,81)

Given a Delaunay
Triangulation of n nodes, How
do I insert node n+1 ?

57
 Delaunay

Bowyer-Watson Algorithm
•Locate triangle that contains
the point
•Search for all triangles whose
X circumcircle contain the point
(d<r)
•Delete the triangles (creating
a void in the mesh)
•Form new triangles from the
new point and the void
boundary
•(Watson,81)

Given a Delaunay
Triangulation of n nodes, How
do I insert node n+1 ?

58
Delaunay

59
 Delaunay

insert point
X

60
 Delaunay

cavity

insert point
X

61
 Delaunay

filled cavity

insert point
X

62
Delaunay

•Begin with Bounding Triangles (or Tetrahedra)

63
Delaunay

•Insert boundary nodes using Delaunay method

(Lawson or Bowyer-Watson)

64
Delaunay

•Insert boundary nodes using Delaunay method

(Lawson or Bowyer-Watson)

65
Delaunay

•Insert boundary nodes using Delaunay method

(Lawson or Bowyer-Watson)

66
Delaunay

•Insert boundary nodes using Delaunay method

(Lawson or Bowyer-Watson)

67
Delaunay

•Insert boundary nodes using Delaunay method

(Lawson or Bowyer-Watson)

68
Delaunay

•Recover boundary
•Delete outside triangles
•Insert internal nodes

69
Delaunay
Node Insertion Methods

Grid Based
•Nodes introduced based on a regular lattice
•Lattice could be rectangular, triangular, quadtree, etc…
•Outside nodes ignored

70
Delaunay
Node Insertion Methods

Grid Based
•Nodes introduced based on a regular lattice
•Lattice could be rectangular, triangular, quadtree, etc…
•Outside nodes ignored

71
Delaunay
Node Insertion Methods

Circumcenter (“Guaranteed Quality”)

•Nodes introduced at triangle circumcenters
•Order of insertion based on minimum angle of any triangle
•Continues until minimum angle > predefined minimum (  30 )
(Chew,Ruppert,Shewchuk)
72
Delaunay
Node Insertion Methods

Circumcenter (“Guaranteed Quality”)

•Nodes introduced at triangle circumcenters
•Order of insertion based on minimum angle of any triangle
•Continues until minimum angle > predefined minimum (  30 )
(Chew,Ruppert,Shewchuk)
73
Delaunay
Node Insertion Methods
A B

Advancing Front
•“Front” structure maintained throughout
•Nodes introduced at ideal location from current front edge
(Marcum,95)

74
Delaunay
Node Insertion Methods

Advancing Front
•“Front” structure maintained throughout
•Nodes introduced at ideal location from current front edge
(Marcum,95)

75
Delaunay
Node Insertion Methods

Edges
•Nodes introduced along existing edges at l=h
•Check to ensure nodes on nearby edges are not too close
(George,91)

76
Delaunay
Node Insertion Methods

Edges
•Nodes introduced at along existing edges at l=h
•Check to ensure nodes on nearby edges are not too close
(George,91)

77
Delaunay
Boundary Constrained

Boundary Intersection
•Nodes and edges introduced where Delaunay edges
intersect boundary

78
Delaunay
Boundary Constrained

Boundary Intersection
•Nodes and edges introduced where Delaunay edges
intersect boundary

79
Delaunay
Boundary Constrained

Local Swapping
•Edges swapped between adjacent pairs of triangles until
boundary is maintained

80
Delaunay
Boundary Constrained

Local Swapping
•Edges swapped between adjacent pairs of triangles until
boundary is maintained

81
Delaunay
Boundary Constrained

Local Swapping
•Edges swapped between adjacent pairs of triangles until
boundary is maintained

82
Delaunay
Boundary Constrained

Local Swapping
•Edges swapped between adjacent pairs of triangles until
boundary is maintained

83
Delaunay
Boundary Constrained

Local Swapping
•Edges swapped between adjacent pairs of triangles until
boundary is maintained
(George,91)(Owen,99)

84
Delaunay

D C
VS

Local Swapping Example

•Recover edge CD at vector Vs

Boundary Constrained
85
Delaunay

E3
D C
E5 E4
E8 E2
E6 E1
E7

Local Swapping Example

•Make a list (queue) of all edges Ei, that intersect Vs

Boundary Constrained
86
Delaunay

E3 E1
D C
E5 E4
E8 E2
E6
E7

Local Swapping Example

•Swap the diagonal of adjacent triangle pairs for each
edge in the list

87
Delaunay

E3
D C
E5 E4
E8
E6 E2
E7

Local Swapping Example

•Check that resulting swaps do not cause overlapping
triangles. If they do, then place edge at the back of the
queue and try again later
88
Delaunay

E3

D C
E5 E4
E8
E6
E7

Local Swapping Example

•Check that resulting swaps do not cause overlapping
triangles. If they do, then place edge at the back of the
queue and try again later
89
Delaunay

D C
E6

Does not maintain

Delaunay criterion!

Local Swapping Example

•Final swap will recover the desired edge.
•Resulting triangle quality may be poor if multiple
swaps were necessary
90
Delaunay
Boundary Constrained
Edge Recovery
DE = edge to be recovered
B •Force edges into triangulation by
performing 2-3 swap transformation

D E

C ABC = non-conforming face

A
3D Local Swapping
•Requires both boundary edge recovery and boundary
face recovery
(George,91;99)(Owen,00)

91
 Delaunay
Boundary Constrained
Edge Recovery
DE = edge to be recovered
B •Force edges into triangulation by
performing 2-3 swap transformation

D E
2-3 Swap
ABCE
ACBD C ABC = non-conforming face

A
3D Local Swapping
•Requires both boundary edge recovery and boundary
face recovery
(George,91;99)(Owen,00)

92
 Delaunay
Boundary Constrained
Edge Recovery
DE = edge recovered
B •Force edges into triangulation by
performing 2-3 swap transformation

D E
2-3 Swap
BAED
ABCE
CBED
ACBD C
ACED

A
3D Local Swapping
•Requires both boundary edge recovery and boundary
face recovery
(George,91;99)(Owen,00)

93
 Delaunay
Boundary Constrained
Edge Recovery
DE = edge recovered
B •Force edges into triangulation by
performing 2-3 swap transformation

D E
2-3 Swap
BAED
ABCE
CBED
ACBD C
ECED

A
3D Local Swapping
•Requires both boundary edge recovery and boundary
face recovery
(George,91;99)(Owen,00)

94
Delaunay

Edge AB to be recovered

Exploded view of tets

intersected by AB

3D Edge Recovery
•Form queue of faces through which edge AB will pass
•Perform 2-3 swap transformations on all faces in the list
•If overlapping tets result, place back on queue and try again later
•If still cannot recover edge, then insert “steiner” point

95
 Delaunay

2-3 Swap 2-2 Swap Face Split Edge Split Edge Suppress

abce, acbd aceb, adcb abce, acbd abnini+1 {i=1…N} abnini+1 {i=1…N}

abde, bcde, cade adeb, edcb abfe, bcfe, cafe abnini+1, nm,knm.jnm.ia,
bafd, cbfd, acfd cbnini+1 {i=1…N} nm,knm.jnm.ib {m=1…M}
N = no. adj. tets at edge ab M = no. unique trias in
polygon P={n1,n2,n3}

96
 Surface Meshing

Direct 3D Meshing Parametric Space Meshing

•Elements formed in 3D using •Elements formed in 2D using parametric

actual x-y-z representation of representation of surface
surface •Node locations later mapped to 3D

97
Surface Meshing

3D Surface Advancing Front

•form triangle from front edge AB

98
Surface Meshing

NC Tangent plane
A
C

3D Surface Advancing Front

•Define tangent plane at front by
averaging normals at A and B

99
Surface Meshing

NC
A
C

D B

3D Surface Advancing Front

•define D to create ideal triangle on
tangent plane

100
Surface Meshing

NC
A
C

3D Surface Advancing Front

•project D to surface (find closest
point on surface)

101
Surface Meshing

3D Surface Advancing Front

•Must determine overlapping or intersecting
triangles in 3D. (Floating point robustness
issues)
•Extensive use of geometry evaluators (for
normals and projections)
•Typically slower than parametric
implementations
•Generally higher quality elements
•Avoids problems with poor parametric
representations (typical in many CAD
environments)
•(Lo,96;97); (Cass,96)

102
Surface Meshing

Parametric Space Mesh Generation

•Parameterization of the NURBS provided by the CAD model can be used to
reduce the mesh generation to 2D

103
Surface Meshing

Parametric Space Mesh Generation

•Isotropic: Target element shapes are
equilateral triangles
•Equilateral elements in parametric
space may be distorted when mapped to v
3D space.
•If parametric space resembles 3D space u
without too much distortion from u-v
space to x-y-z space, then isotropic
methods can be used.
v

104
Surface Meshing

Parametric Space Mesh Generation

•Parametric space can be “customized” or warped so that isotropic methods can be used.
•Works well for many cases.
•In general, isotropic mesh generation does not work well for parametric meshing

u
v

u
Warped parametric space

105
Surface Meshing

Parametric Space Mesh Generation

•Anisotropic: Triangles are stretched based on a
specified vector field
•Triangles appear stretched in 2d
v
(parametric space), but are near equilateral
in 3D
u

106
Surface Meshing

Parametric Space Mesh Generation

•Stretching is based on field of surface derivatives
  u  u u   v v v 
u   , ,  v   , , 
  x y z   x y z 
•Metric, M can be defined at every location on  u u u 
v
surface. Metric at location X is:  , , 
 v v v   x y z 
u  x , y , z 
E F   
M (X)   
F G y
x
z
E  u  u F  u  v G  v  v

107
Surface Meshing

Parametric Space Mesh Generation

•Distances in parametric space can now be measured as a function of direction and
location on the surface. Distance from point X to Q is defined as:

l ( XQ)  XQT M ( X) XQ

v Q
M(X)
X Q u X l ( XQ)
l ( XQ)
v y
x
z
u

108
Surface Meshing

Parametric Space Mesh Generation

•Use essentially the same isotropic methods for 2D mesh generation, except distances
and angles are now measured with respect to the local metric tensor M(X).
•Can use Delaunay (George, 99) or Advancing Front Methods (Tristano,98)

109
Surface Meshing

Parametric Space Mesh Generation

•Is generally faster than 3D methods
•Is generally more robust (No 3D
intersection calculations)
•Poor parameterization can cause
problems
•Not possible if no parameterization is
provided
•Can generate your own
parametric space (Flatten 3D
surface into 2D) (Marcum, 99)
(Sheffer,00)

110
Smoothing
P4

P3
P

P5

P2

P1
(Field, 1988)
Laplacian
111
Smoothing
P4 Centroid of
attached nodes

P3

P
P5

P2 n
 Pi
P1
P  i 1
(Field, 1988) n
Laplacian
112
Smoothing
Centroid of
P4 attached nodes
Can create
inverted elements

P
P5
P1

P3
n
 Pi
P6 P  i 1
P2 n
Laplacian
113
Smoothing
Centroid of
P4 attached nodes
Can create
inverted elements

P5
P1

P3
P
n
 Pi
P6 P  i 1
P2 n
Laplacian
114
Smoothing
Centroid of
P4 attached nodes
Can create
inverted elements

P
Constrained Laplacian: Do not
P5 move node unless minimum
P1 distortion metric is improved

P3
n
 Pi
P6 P  i 1
P2 n
Laplacian
115
Smoothing

Smoothing Procedure
LET iter=0
SET all NODE v to active
REPEAT (main smoothing loop)
FOR EACH NODE v DO:
1. If v has been deactivated from smooth, then CONINUE.
2. distance moved = move_node(v)
3. If distance moved < move tolerance, then
deactivate v
Else,
Allow move.
If neighbor node to v is inactive, then activate it.
END DO
iter++;
UNTIL all nodes are deactivated OR iter > MAX_ITER (end of main smoothing loop)

116
 Smoothing

Optimization- min Compute shape

Based y  metric i for each

quad at P
 min
Method
P x Choose min to
 optimize

Steepest descent
optimization Compute gradient
 vector g=gmin by
y perturbing small 
in x and y

x
(Canann, Tristano,
  i   min 
Staten, 1998) Pnew  P  g   min  
 g  g  g  g i 

117
 Smoothing

Compute shape
Optimization- 
min

metric i for each
Based y quad at P

Method min
x  Choose min to
optimize

Steepest descent
optimization Compute gradient
 vector g=gmin by
y perturbing small 
in x and y

x
(Canann, Tristano,
  i   min 
Staten, 1998) Pnew  P  g   min  
 g  g  g  g i 

118
 Smoothing

Compute shape
Optimization-
 metric i for each
Based quad at P
 min
Method y
Choose min to
 min
 optimize
 x
Steepest descent
optimization Compute gradient
 vector g=gmin by
y perturbing small 
in x and y

x
(Canann, Tristano,
  i   min 
Staten, 1998) Pnew  P  g   min  
 g  g  g  g i 

119
Smoothing
Triangle Distortion Metric
C
NS = Surface normal
area ( ABC )
 ( ABC )  ( I ) 3 2 2 2
l1  l2  l3
l2
1 for CA  CB   N S  0
I
l3 1 for CA  CB   N S  0
B C
C
C

l1 B
A
A B B A
(Canann, Tristano, Staten, 1998) A
=1 =0 =-1

120
Smoothing

C
Quad Distortion Metric
D
Compute four triangle metrics
 (ABC )  (CDA)
 (BCD )  (DAB )
B Order metrics in descending order
1   2   3   4
A  3 4

1 2

(Lee, Lo, 1994)

121
Smoothing
Combined Laplacian and Optimization
Based Smoothing
Compute initial distortion metrics for all elements
LET iter=0
SET all NODE v to active
REPEAT (main smoothing loop)
FOR EACH NODE v DO:
1. If v has been deactivated from smooth, then CONINUE.
2. If v wasn’t moved by optimization-based smoothing in last iteration, then
distance_moved = move_node_laplacian(v)
3. If (iter >= 2) then
distance_moved = move_node_optimize(v)
4. If distance_moved < move_tolerance AND min > min_allowable_metric, then
deactivate v
Else,
Allow move, and update adjacent distortion metrics to v.
If neighbor node to v is inactive, then activate it.
END DO
iter++;
UNTIL all nodes are deactivated OR iter > MAX_ITER (end of main smoothing loop)

(Canann, Tristano, Staten, 1998)

(Freitag, 1997)

122