Boris N.

Delaunay
 MOTIVATION: TERRAIN
TRIANGULATION
Property of triangulation
SOLUTION OF TERRAIN PROBLEME
ANGLE OPTIMAL TRIANGULATION
EDGE FLIPPING
THALES THEOREM
ILLEGAL TRIANGULATION
LEGAL TIANGULATION
DELAUNAY GRAPH
Property of Delaunay Graph
DELAUNAY TRIANGULATION
Property of Delaunay Triangulation
Computing Delaunay Triangulation
RANDOMIZED INCREMENTAL CONSTRUCTION
Application of Delaunay Triangulation
Euclidean Minimum Spanning Tree
Euclidean Traveling salesperson Problem
REFERENCES
 We can model a piece of the earths surface as a terrain.
A terrain is a 2-dimensional surface in 3-dimensional space with a
special property: every vertical line intersects it in a point, if it
intersects it at all. A terrain can be visualized with a perspective
drawing like the one in bellows Figure

Figure: A perspective view of Terrain

 To build a model of the terrain surface, we can start with a
number of sample points where we know the height.

Of course, we dont know the height of every point on earth; we only

know it where weve measured it.
This means that when we talk about some terrain, we only know the
value of the function f at a finite set P A of sample points.
From the height of the sample points we somehow have to
approximate the height at the other points in the domain.
A native approach assigns to every p A the height of the nearest
sample point. However, this gives a discrete terrain, which doesnt
look very natural.

Figure: A terrain from a set of sample points

So, How can we most naturally approximate height
of points not in A?

Determine a
Triangulation of A in
R, then raise points to
desired height
Triangulation: planar
subdivision whose
bounded faces are
triangles with vertices Figure : Triangulation
from A
Maximal planar subdivision: a subdivision S such that
no edge connecting two vertices can be added to S without
destroying its planarity

Plane Graph: No two edges in the embedding cross.

Triangulation of set of points P: a maximal planar

subdivision whose vertices are elements of P.
 Outer polygon must be
convex hull that is the
union of the bounded faces
of T is always the convex
hull of P, and that the
unbounded face is always
the complement of the
convex hull.

Internal faces must be

triangles, otherwise they
could be triangulated
further. Convex Hull
Here, Some triangles are better than other in triangulation.
SO, now remain a question that-
But which triangulation is the most appropriate one for
our purpose, namely to approximate a terrain??
There is no definitive answer to this question. We do not know the
original terrain, we only know its height at the sample points.
Since we have no other information, and the height at the sample
points is the correct height for any triangulation, all triangulations
of P seem equally good. Nevertheless, some triangulations look
more natural than others.

So here we will avoid skinny triangles. That means we will try

maximize the minimum angles which is basically DELAUNAY
TRIANGULATION
SO, TERRAIN Problem solve by Angle Optimal
Triangulation(Edge Flipping) and Delaunay Triangulation
 To create a angle vector Let T be a triangulation of P with m
triangles. Its angle vector is A(T) = (,,,.....,m) where
,,,,m are the angles of T sorted by increasing
value.
Let T is another triangulation of P. Its angle vector is A(T)
= (,,,.......,m) Then we say the angle-vector of T is
larger than the angle-vector of T if A(T) is lexicographically
larger than A(T). This means iff there exists an index i with
1 i 3m such that j=j for all j<i and > then
A(T)> A(T).
So , Best triangulation is triangulation that is angle optimal,
i.e. has the largest angle vector. That is Maximizes
minimum angle. Then we say the triangulation T is angle-
optimal if A(T) A(T).
.
 If the two triangles form a convex quadrilateral, we could have an
alternative triangulation by performing an Edge Flip on their shared
edge.
In the Edge Flipping just change diagonal or an edge which shared
by two triangles and also change angles.

Edge Flip

Change the angle vector (,,) replaced by ,,)

The edge e = PiPj is Illegal Edge if min 1i6 < min 1i6
Flipping an illegal edge increases the angle vector
 We can use Thales Theorem to test if an edge is legal
without calculating angles
Denote the smaller angle
defined by three points p, q,
r.
Let C be a circle, l a line
intersecting C in points a and
b and p, q, r, and s points
lying on the same side of l.
Suppose that p and q lie on C,
that r lies inside C, and that s
lies outside C. Then:
 If triangulation T contains an illegal edge e, we can
make A(T) larger by flipping e.
In this case, T is an illegal triangulation.
The edge PiPj is illegal if and only if Pl lies in the
interior of the circle C.
C
Pl

Illegal Edge
 A legal triangulation is a triangulation that does not contain
any illegal edge.

Algorithm LegalTriangulation(T)

Input. A triangulation T of a point set P.

Output. A legal triangulation of P.
1. while T contains an illegal edge pipj
2. do ( Flip PiPj )
3. Let PiPjPk and PiPjPl be the two triangles adjacent to pipj.
4. Remove PiPj from T, and add PkPl instead.
5. return T
Flip illegal edges of this triangulation until all edges are legal . Algorithm
terminates because there is a finite number of triangulations of input points
P.
The strait line embedding of a graph G, which has a line
connecting two nodes if their corresponding Voronoi cells
share an edge, denoted as Delaunay Graph of P ie. DG(P).

DG(P)
 No two edges cross; DG(P) is a planar graph.
PiPj is in DG(P) iff closed disc Cij Contained in v(Pi)
with Pi and Pjon the boundary and no
other site is contained in it.
Let tij be the triangle formed by Pi,Pj
and center of Cij.
Note that edge of tij between Pi and
center of Cij is inside V(Pi).
Let another edge PkPl also in DG(P)
such that PiPj and PkPl
intersect.Define Ckl and tkl for PkPl.If
intersected PiPj, it must also intersect
one other edge e of tij.Because Pk Contained in v(Pi)
and Pl are outside Cij and therefore outside tij.Say an edge e of tkl must
intersect PiPj.
Notice that one of the edges of tij incident to center of Cij and one of the edges
of tkl incident to center of Ckl must intersect.
But, those edges must be contained within their respective Voronoi cells,
which is a contradiction.
If the point set P is in general position then the Delaunay
graph is a triangulation.

General Position: We say that a set of points in general position if it

contains no four points on a circle. Therefore, P in general position means
every vertex of the Voronoi diagram has degree three, and consequently
all bounded faces of DG(P) are triangles.

DG(P)

VOR(P)

Delaunay Triangulation
Let P be a set of points in the plane, and let T be a triangulation of P. Then
T is a Delaunay triangulation of P if and only if the circumcircle of any
triangle of T does not contain a point of P in its interior.If the circumcircle
of any triangle does not contain other points of P then its called Empty
Circle property.
Let P be a set of points in the plane. A triangulation T of P is
legal if and only if T is a Delaunay triangulation of P.
We shall prove that any legal triangulation is Pm
Delaunay Triangulation by contradiction.So assume
T is a legal triangulation of P that is not Delaunay
Triangulation.As T is legal so adjacent edge of Pl
triangles is legal.As T is not Delaunay Triangulation Pi
so there is a PiPjPk ,the circum circle C(PiPjPk) Pj
contains a point Pl P in its interior. Let e=PiPj be
the edge of pipj pl such that the triangle PiPjPl does
not intersect PiPjPk Now look at the triangle pipj pm
adjacent to pipj pk along e. Since T is legal, e is
Pk
legal.so Pm does not lie in the interior circle of
C(PiPjPk).The citcum circle C(PiPjPm) of PiPjPm contains the part of C(PiPjPk), separeted
from PiPjPk by e.Consequently Pl C(PiPjPk).Assume that edge PjPm is the edge of the
PiPjPl so this is not intersect PiPjPm.But by Thales Theorem, PjPlPi > PjPmPi.
a contradiction as it allows PjPm to be flipped.
Let P be a set of points in the plane. Any angle-optimal triangulation of P
is a Delaunay Triangulation of P. That is any Delaunay triangulation of P
maximizes the minimum angle over all triangulations of P.

min = 27 After edge flip for min = 50

angle optimal
D triangulation D

C C
90
A A

B B
 If we can verify for every edges that the two triangles are sharing the
edges are legal then it if Delaunay Triangulation.
We cant take every circum circle and check the respectively point.
But we only look at the edge which is sharing the two triangles and
draw the circum circles if they are not contain any other point and
this is holds for every pair of trianngles then its a Delaunay
Triangulation. This is called Local Property
 A simple algorithm to compute Delaunay Trianguletion. It is Randomize
Incremental Construction
1. This algorithm start with a big triangle say P-1P-2P-3 that contains all the points.
The vertices of this big triangle should not be lie in any circle defined by 3
points in P.
2. Add one point Pr at a time, then add edges from Pr to the verticies of the
existing triangle.
3. Then two cases arise here:
Pr lies in the interior of the triangle. (figure: A)
Pr falls on an edge (need to make sure new edges are legal by flipping
edges if necessary). (figure B)
Pi
Pk

Pl Pr Pk
Pi Pr
Pj Pj
Input: A set P of n points in the plane
Output: A Delaunay triangulation of P

1. Let p-1 , p-2 & p-3 be 3 point s.t. P is contained in triangle p-1 p-2 p-3
2. Initialize T as triangulation consisting of a single triangle p-1 p-2 p-3
3. Compute a random permutation p1 , , pn of P
4. for r 1 to n
5. do (* Insert pr into T : *)
6. Find a triangle pi pj pk T containing pr
7. if pr lies in the interior of the triangle pi pj pk
8. then Add edges from pr to 3 vertices of pi pj pk and split it
9. LegalizeEdge(pr , pi pj , T)
10. LegalizeEdge(pr , pj pk , T)
11. LegalizeEdge(pr , pk pi , T)
12. else (* pr lies on an edge of pi pj pk, say the edge pi pj *)
13. Add edges from pr to pk and to the third vertex of
other triangle that is incident to pi pj , thereby
splitting 2 triangles incident to pi pj into 4 tris
14. LegalizeEdge(pr , pi pl , T)
15. LegalizeEdge(pr , pl pj , T)
16. LegalizeEdge(pr , pj pk , T)
17. LegalizeEdge(pr , pk pi , T)
18. Discard p-1 , p-2 and p-3 with all their incident edges from T
19. return T
 Algorithm_Legalize_Edge(pr , pi pj , T)
1. (* The point being inserted is pr , and pi pj is the edge of T
that may need to be flipped *)
2. if pi pj is illegal
3. then Let pi pj pk be triangle adjacent to pr pi pj along pi pj
4. (* Flip pi pj *) Replace pi pj with pr pk
5. LegalizeEdge(pr , pi pk , T)
6. LegalizeEdge(pr , pk pj , T)
This meet the Delaunay Triangulation

Pk But,

Pr Pj

Pr
Pi

Here after adding this Pr, the circum circle of this intirior point contain
some other point of set P.So,Now flipping the edge for restore the
Delaunay Property.
Pr

After flipping we get new two tryangle

and draw the circum circle and check
the Delaunay Property,but this figure
again not meet the Delaunay
Property.So we flipping the edge again.
After flipping edge we check again
Delaunay Property and this
triangulation meet the Delaunay
property because the ciecum circle is not
contain ny othe point. So, Now this
triangulation is Delaunay
Triangulation.
Pr

This flipping is continue until the triangulation becomes legal

and every time flip only the diagonal one of the end point of
the diagonal is the new point and then the previous triangles
are legal so it must be include the new point
 In the Algorithm of Randomize Incremental Construction , find the
triangle which containing Pr point.
While we build the Delaunay triangulation, we also build a point
location structure D. The leaves of D correspond to the triangles of
current triangulation T. We maintain cross pointers between the leaves
and the triangulation.
Each node in D corresponds to triangle that was created by some point.
We initialize D as a DAG(Directed Acyclic Graph) with a single leaf
node that corresponds to the big triangle.
Searching for a point Pr would be going down the DAG D through a
sequence of nodes corresponding to all triangles created before step r
that contain Pr.
Whenever we split a triangle 1 into smaller triangles a and b (and
possibly c), add the smaller triangles to D as leaves of 1
For the rest, see the pictorial example.


Split
B C
Pj

A Pr

A B C
Pi
 Flip PiPj
B
Pj

A B C
A Pr Pk

4 5
Pi

Flip PiPk

B
Pj
A B C

A
Pr Pk


Pi Pl

Let M be the maximum value of any
coordinate of a point in P. Then , big
the triangle has the vertices at
P-1 = (3M, 0),
P-2 = (0, 3M)
and
P-3 = (- 3M, - 3M).
This are not actual coordinate.

M
M
The Randomized Incremental Algorithm for the Delaunay
Triangulation of n points takes O(n log n) time and O(n)
space, both on expectation.
From nodes in D representing triangles created .Each triangle takes O(1) time, so
the total time will be O(n) + (time for point locations).
Now Searching for a point Pr would entail going down the DAG D through a
sequence of nodes corresponding to all triangles created before step r that
contain Pr.
So each triangle that was created is visited once for each point in it. In other
words, . . . Let K() P be the points inside triangle . Time for point locations is

K()

each triangle created in round r has O(n/r) points of P in it. Therefore,

K() O(n/r)

O n(log n)
 Euclidean Minimum Spanning Trees
For a set P of n points in the plane, the
Euclidean Minimum Spanning Tree is the
graph with minimum summed edge length
that connects all points in P and has only
the points of P as vertices

The Euclidean Minimum

Spanning Tree does not have
cycles.
Suppose G is the shortest connected
graph and it has a cycle. Removing one
edge from the cycle makes a new graph Euclidean Minimum
G0 that is still connected but which is Spanning Tree
shorter. Contradiction
 Every edge of the Euclidean Minimum Spanning Tree is an
edge in the Delaunay graph

Suppose T is an EMST with an

edge e = pq that is not Delaunay
Consider the circle C that has e as
its diameter. Since e is not
Delaunay, C must contain another P
point r in P. r
Either the path in T from r to p e
passes through q, or vice versa. T
T
The cases are symmetric, so we
can assume the former case.
Then removing e and inserting pr q
instead will give a connected
graph again (in fact, a tree)
Since q was the furthest point
from p inside C, r is closer to q, so
T was not a minimum spanning
tree. Contradiction
Which is not
possible
A data structure exists that maintains disjoint sets and allows the following two
operations:
Union: Takes two sets and makes one new set that is the union.
Find: Takes one element and returns the name of the set that contains it
If there are n elements in total, then all Unions together take O(nlogn) time and each
Find operation takes O(1) time
Let P be a set of n points in the plane for which we want to compute the
EMST

1. Make a Union-Find structure where every point of P is in a separate set

2. Construct the Delaunay triangulation DT of P
3. Take all edges of DT and sort them by length
4. For all edges e from short to long:
Let the endpoints of e be p and q
If Find(p) Find(q), then put e in the EMST, and
Union(Find(p),Find(q))
So, The Euclidean Minimum Spanning Tree of P can be computed in O(nlogn)
time
 Given a set P of n points in the plane, the Euclidean Traveling
Salesperson Problem is to compute a tour (cycle) that visits all points of P
and has minimum length . A tour is an order on the points of P (cyclic
order). A set of n points has (n-1)! different tours

We can determine the length of each tour in O(n) time: a brute-force

algorithm to solve the Euclidean Traveling Salesperson Problem (ETSP)
takes O(n) O((n-1)!) = O(n!) time

If an algorithm A solves an optimization problem always within a factor

k of the optimum, then A is called an k-approximation algorithm.

If an instance I of ETSP has an optimal solution of length L, then a k-

approximation algorithm will and a tour of length k L
We will use the EMST to approximate the ETSP

proceed
start
follow
skipping
toand
getan
closethis
toatedge
any way
vertex
visited
theon
another vertices
one side
tour
vertex
 The walk visits every edge twice,
so it has length 2EMST
The tour skips vertices, which
means the tour has length
Pi
2EMST
The optimal ETSP-tour is a
spanning tree if you remove any
edge!!!
So EMST < ETSP
Remove the edge PiPj . So now this
Pj
ETSP is EMST.

A tour visiting all points whose length is at most twice the

minimum possible can be computed in O(nlogn) time. So an
O(nlogn) time, 2-approximation for ETSP exists

As ETMST is meet the Delaunay property and evry edges are Delaunay
edges then ETSP is also meet Delaunay Property and its edges are
Delaunay edges
DOCUMENTS:

E:\PG\SEM 3\SEMINER\Delaunay2D.ppt

E:\PG\SEM 3\SEMINER\Computing the Delunay Triangulation.ppt

E:\PG\SEM 3\SEMINER\interpolation.pdf

E:\PG\SEM 3\SEMINER\DELAUNAY TRIANGULATION.mp4

WEBSITES

https://en.wikipedia.org/wiki/Delaunay_triangulation

http://www.cs.uu.nl/docs/vakken/ga/slides9alt.pdf

http://www.cse.iitm.ac.in/~augustine/cs6100_even2012/slides/09_
DelaunayTriangulation.pdf

https://www.youtube.com/watch?v=IqdSdbxrTsY