CS 763 Computational Geometryalubiw/CS763/CS763-Lecture9-final.pdf · (nextslide. Lecture 9:...
Transcript of CS 763 Computational Geometryalubiw/CS763/CS763-Lecture9-final.pdf · (nextslide. Lecture 9:...
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Voronoi diagram
Given points P = {p1, . . . , pn} in the plane, the Voronoi region of pi is
pi is called a site.
The Voronoi diagram consists of all the Voronoi regions
To see that the Voronoi diagram consists of straight line segments:
CS763-Lecture9 1 of 23
.
.
- i
•.
- i-- in :
-
.
, noise(
(
:
Hctigpj ) - points closer totithanpj it .
-_ half-plane pi. t.tt -
i
✓ Epi)=j±?HCpi , pj ) i t%p;HE,pj)
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
CS 763 Computational GeometryPost Office Problem: find the closest post office. O(n) ?
Application: Post Office Problem
Given n points in the plane (post offices) preprocess to handle query: which post office is closest to query point p
Voronoi Diagram
CS 763 Computational GeometryPost Office Problem: find the closest post office. O(n) ?
Compute the Voronoi diagram.Query becomes: which region contains p ?This is Planar Point Location.
— also closest airfield for airplane in flight.
Recall
CS763-Lecture9 2 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
History
Georgy Voronoy, 1908, Russian/Ukrainian mathematician
but Voronoi diagrams were used before that:- Descartes 1644- Dirichlet 1850, “Dirichlet tessellation” - used to prove
unique reducibility of quadratic forms
also used in- crystallography - crystals growing outward from “seeds” form Voronoi regions- epidemiology, 1854, mapping cholera by proximity to infected water pump in London- geography and meteorology - “Thiessen polygons” 1911- condensed matter physics - “ Wigner–Seitz cells”- natural sciences - “Blum’s transform” = medial axis of polygon, used to describe shapes- aviation, to identify nearest airfield
dual structure: Delaunay triangulation
Boris Delone (but used French transliteration “Delaunay”), 1934
https://en.wikipedia.org/wiki/Voronoi_diagram
CS763-Lecture9 3 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Descartes 1644
CS763-Lecture9 4 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Some nice illustrations, including Voronoi diagram of moving points
http://datagenetics.com/blog/may12017/index.html
http://datagenetics.com/blog/may12017/anim2.gif
https://www.youtube.com/watch?v=z3yy5aBv0ug
CS763-Lecture9 5 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
http://i.stack.imgur.com/gcajM.gif
CS763-Lecture9 6 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Voronoi diagram of US Starbucks
Jim Davenport
CS763-Lecture9 7 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Voronoi diagram of capital cities
CS763-Lecture9 8 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
http://www.flong.com/projects/zoo/
CS763-Lecture9 9 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
http://blog.kleebtronics.com/voronoi/
CS763-Lecture9 10 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Terminology site
Voronoi vertex
Voronoi edge
Properties
- empty circle property
- degrees of Voronoi vertices?
- shape of Voronoi cells?
- when is a cell unbounded?
- number of vertices/edges/cells
Voronoi cell
CS763-Lecture9 11 of 23
Con next slide)
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Terminology site
Voronoi vertex
Voronoi edge
Properties
- empty circle property
- degrees of Voronoi vertices?
- shape of Voronoi cells?
- when is a cell unbounded?
- number of vertices/edges/cells
Voronoi cell
empty circle = no sites inside
CS763-Lecture9 12 of 23
Voronoi vertex thefor each stain, makeancircle centered at vertexthrough the 3 Cor more? I equally close sitesThen this circle has no other site inside it\ degree =3 assuming no 4 co -circularpoints .
\convex polygons (may be unbounded )
} on next sides\ # cells =#sites = n
A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Properties of Voronoi diagram
V(pi ) is unbounded iff pi is on the convex hull of the sites
There are ≤ 2n Voronoi vertices and ≤ 3n Voronoi edges
CS 763 F20
CS763-Lecture9 13 of 23
÷:
missin":c::y÷÷÷÷÷÷"'
iii. insideinf . ray invcpi) ' .Asx -70 on the ray , get empty
half-planeEuler :r-etf-zf-n-ftsites.fiNote : deal with infinite edges by adding a'
'
point at
, infinity" where all infinite edges go .
( every vertex has degree 33i t '
i s 2e=Edegr 230I s
to '
s
- 2=v- etf Ev - EwtnEpt at infinity vs In -27
CE 3cm - 2) .
A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Delaunay triangulation
The Voronoi diagram can be captured by a purely combinatorical structure (versus computing coordinates of Voronoi vertices)
Given points P = {p1, . . . , pn } in the plane, the Delaunay triangulation isa graph with vertices p1, . . . , pn and edge (pi , pj ) iff V(pi ) and V(pj ) share an edge.
is the planar dual of
CS 763 F20
Voronoi edgesDelaunay edges
siteNote: a Voronoi edge and its corresponding Delaunay edge do not always cross.
CS763-Lecture9 14 of 23
A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Delaunay triangulation
CS 763 F20
Properties
- it is a triangulation
- edge/face iff empty circle through those sites
- what is the boundary ?
Voronoi edgesDelaunay edges
site
CS763-Lecture9 15 of 23
(next slide
A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Delaunay triangulation
CS 763 F20
Properties
- it is a triangulation
- edge/face iff empty circle through those sites
- what is the boundary ?
Voronoi edgesDelaunay edges
site
CS763-Lecture9 16 of 23
✓ Delaunay triangulation
.
:c:::::'
namesima :::÷.
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Properties of Delaunay triangulations
(pi , pj ) is an edge iff there is an empty circle through pi pj
Proof
CS763-Lecture9 17 of 23
-- -
n
pi i n
Cpi ,p;) is an edge of DCP)
i•*. I
if vcpi) and V Cpj) share i i,
>C l i
a boundary points ( using no not - --
4 co- circular) Pj
if the circle centered atx through pi and pjis empty ( has no other sites )
DX
⑤ Ya'm't Vusi)happen Yf⇒Kpi
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Properties of Delaunay triangulations
No two Delaunay edges cross
Note that this must be proved — it is not true for general planar maps
Proof
CS763-Lecture9 18 of 23
Suppose cu , v) and Gc , y)
care Delaunay edges that cross
u ••① 4479 are sites ,÷:G- Consider circles through u and ✓
One , say C ,is empty (by previous prop .)
where are a and y ?Must lie outside C and Ky crosses urso one to either side
claim : then # an empty circle though x andy .
Needs a bit more proof . DX
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Properties of Delaunay triangulations
Sites form a face iff there is an empty circle through themi.e., (since we assumed no 4 sites on a circle)pi pj pk form a face iff there is an empty circle through them
Proof
CS763-Lecture9 19 of 23
Suppose there is an empty circle BE-
'
-
⇐ M'
By above we have ff\3 edges pi Pj pjpkpkpi ATE
and nothing crosses them*s'.- -
-
t Pk
so this d is a face
⇒Suppose pistz.pk form a face .
pi.• bgkia.fm?iirtdeisthrgpgs.h3 points*
q No site inside d.(because it is a face>Pi can there be a site q ?
No , because otherwise there isno empty circle through pi and pj .
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Properties of Delaunay triangulations
The boundary of the Delaunay triangulation is the convex hull of the sites
Proof
CS763-Lecture9 20 of 23
ideae= cu ,
o) is an edge of CH
?! . . . . . . , if perpendicular bisector•
°
!! .'
'
iq
is unbounded Vor . edgea a @
( v int. large empty circles
i. through u and riff no o in Dlp) on thatside of ur
iff Cup is boundary edgeof triangulation .
A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
An alternative definition of Delaunay triangulation:
for each empty circle through 3 points, add a triangle
(part of)
CS 763 F20
Delaunay triangulation
CS763-Lecture9 21 of 23
withoutgoing via Voroui diagram
A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
Application of Delaunay triangulations: finding all nearest neighbours
Given n points in the plane find, for each point, its nearest neighbour — gives nearest neighbour graph, a directed graph of out-degree 1.
Many applications, e.g.
in statistical analysis:find hierarchical clusters usingnearest neighbour chain algorithm
https://demonstrations.wolfram.com/NearestNeighborNetworks/
CS 763 F20
More on this in the next lecture.
CS763-Lecture9 22 of 23
CS 763 F20 A. Lubiw, U. WaterlooLecture 9: Voronoi Diagrams
References
- [CGAA] Chapters 7, 9
- [Zurich notes] Chapters 5, 7 (they start with Delaunay)
- [O’Rourke] Chapter 5
- [Devadoss-O’Rourke] Chapter 4.
Summary
- Voronoi diagram and Delaunay triangulation definitions, relationships, properties
CS763-Lecture9 23 of 23