0% found this document useful (0 votes)
12 views

Computer Vision Module-3 Notes

The document discusses various techniques for image segmentation, focusing on edge detection methods such as Sobel and Canny algorithms. It highlights the importance of detecting edges based on abrupt changes in pixel intensity and outlines the mathematical foundations behind these techniques. Additionally, it covers concepts like thresholding and the use of masks in identifying and processing edge points in images.

Uploaded by

manumanusha298
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
12 views

Computer Vision Module-3 Notes

The document discusses various techniques for image segmentation, focusing on edge detection methods such as Sobel and Canny algorithms. It highlights the importance of detecting edges based on abrupt changes in pixel intensity and outlines the mathematical foundations behind these techniques. Additionally, it covers concepts like thresholding and the use of masks in identifying and processing edge points in images.

Uploaded by

manumanusha298
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 25

LMAGE SEGMENTATION:

ion Bub dirde


còto i
Image 8egmentat
Constilaentregona
approatha
Thre fundamtntally lioo kinds af
Dis Contimulg * Similavilj
kearnentaton:
loilymey be due to plxl tbttnai
Smi
Colouror le Lute
-Die contnily.à due to Ad Change
den

y, colos (exture.
ch pixelinten2i
o

There ae Ehee kinde o} DiconH Ohe o

Teolalespointz

and Eog
vie wed
point be o

line whose
lengn eqas to
e
pixel
bo Vte wed a an
may edge kagment
tn huch
y
tntenai oy tauLoround on
eulhen i de ne 'ei
tiers mub
múch louey than
hugher
the

% on
Lune

cage
pxels

pria
unton
t ohich the
abrupty
chonge
chtnil

keh q Connelrd edg pxee. pri.


The Common lo took ar decontiuthee
Bmeee mosk over the mag
The mask ddelermna
Cook fo :
which kind fducont -

313

Deletion Teolaladpoinla
the Koplaulan
|UMng

ohere

(2 )

This expreLon cun be nplemnledUng


be exended to ncude the
duagona
Ea (3) Can belo
shoon
terme and

(-8 - - (A)

gaud lo be
mak (A) point à

deteced
the

t ohich he

reeponae
mosk e Cenerad

the moak
absola value o he

at thot point exceeda Apeye threheLd.


&uch are locbelledI output
ponl
othera labelledO, hu rodcng
bunom iag
Nothermatiealy
|Re*y)T.
0herwlee

Line Deleeton
he
For Digctol mage threepoint straugt
i
Hori2onlil Verticad oY

CuigonasCt4g°or- ys').
The kne deltonmole ae aa Rhoon belou

22. 2
2 --
2 2 2
2 2 2
Hos 2ontal Vertal
i
R R2, Rs and Ru denole the reapon%e g -0
Let
he
mos KS trom lett6 ght

wheYe Rspon2e R =

Ke|

Suppo2e that rnage * HLteredCinduvnduolly


oilh the -four maskB'

aee ]k,
point(n the mage
gcven
that pont i'a soud to be more
noaled oil5 a tò th lre ction o

takely
mask K.

E DGE Delection
delietionta trmase
based u C hangs
abrupt local)
to
accordinig
intenil
model clauted
proyilea
a Eranaitionbetween
Ovr the
cdeally
tuo tinhnailg devea oceureng
RAMP EDGEE
To pTactite
hat
duqtal rhaqa houwe
ave blured and me ay Tn Such ilubdions
edqu
more modelled o hang
etoady
thtenailiyRamp protlean shoorn tn tgue

model,an
edge pointà eng point
Contauned the Ramp and
n
ede kegment
$ edae poinlathatane-Conneted-.

Roo Edae:
modeu

Too edge beng deteruned


by the hickne k Shapnee Une
has he Charaeters
toA
t-qure
Consdus the 6hoon blous

-torntonta
thtene

Fret
duunc!

Reeond
Du vtie

yt to might aleng inttnag


Movinig rom progle,
he 4otoong ponlican be noled:
Derivatveu poHve at (he
Yamp
*) Second dervatveú postve at Fe
ramp at end ram 4 ,zero at poinR
on
negalave
ramp
hrom hee Obeervatio,
Used to
Can be

delicttha pre lenta og edge.


Uied deenne
Beond derivatt
ve Can be
G

hether Plxel es on he
edge

lkqut
to noue, he

Snce bolh he deIvativea are eeneite


ane
alape per torned in edqe dalieton
three fundamenlal
SrooothIn tor noiie Reduetion:
Image

image ate that are


- Cxlrai fromn poinla
come edqe poinla
polentai candidaliyoa to be
localiathon: the Landidalz
Seleelsfrom

edge pon only the poinla


ponla thatae brue membre

an
5 the Bet o pont Compisng edqe

Basic Edqe Deltoe.


FYst Order Drivatives- Image raduen

- The qvadient g
an at
image ftry)
ocoton

Or Vector:

Magnilade thie :
Vectos

Dineettoo thiu Vecto:


aiona an be
Linplernenled by
mece tcoo guaie

btegest a
when Dia edg

Consider the
3x3

Zu 2s 2c

Roberte CYO- qvadient opatore

Pre-oiteoperatore -

-)
Sobel opurator

OE. (K+9Zg+2g)-CZ1+Z+23)

-lo

th the Center locothon provdu tmege

smoothin

Nete i- Dreuoit moke kmpler to tinplement

4han sobe mas.


* Sobl masta heve bettey nouie- luPpreMbn

Charoeterstu
al moelum to Zero)
Phus

Conalant. nttnilg

pouuble .to tho 3x3 mote


modiby
bo that have thu ktrongutTuponte arg
ogonel durections. itt Sobel moole
dieetona
needed tor detcttngedqts th daagonal
hoor
1
12 -2-o

-)-) -2-) t2
prewitt Sobel

Delietion.
IMore Ad vanced Technq or Edqe

Ma- HLdveth edoe beteclos

Consder G a-D Gauuan functio


n
Gty) e
StandNd duv(aton

Colecting term ve the fnalexpron:


e
[
CLan
Called the dapla
a
Thi epren on

Gaunan (Lo).
Maske orY Ae eon be qeneraled by
arbitr
Sannpleng the LoG cquaton .
a) 3-D plot% the

G
an
inag
Shouoin 2ero

(d)5x5 mask
0 0
Oup pou matib
-2 -1 to Bhonpe in ):
0

-1

Zero crossing
-1 -2 -1 0

-Zeroerosing

2V2o
0 -)0 0

(d)

Tbe Mo- Hldreth algornthrs ConAuR Conrebrng

LoG fiLterwith Cn inputimag 4xy

4 then fndintthe Zero


to determne the locatibn o edqs
unee above
prou
Can be 0Uen as.

as-hlethnoled hot itig porbk to apporal


wil
(DO):-

DoGCy
The Canny edqe Dleator
Cannys pproach ia bas ed on threebaxi
ARl
edgee athouldbe found,
k
4hrere shuld be no $purioue Tepongee
9- Edqe ponta &houtd be oee locai3ed:
The edge lbcoed muat be as cloeao
pocble to
3
Single ed ge point apone : The delector
&hould
geturn One pointfor each true edge Pont
Srnooth the put inag
Let f(xy denola the
he
tinputtinaqe K G(y)denole
qou an tuncton:

forna Bmoohed

a Compule the gradntmagnilade & ange imaga

ton

to obtaun
*nyt tltermauk pair can be Uged
3 non- manmd Buppreeon lo gnradent

magniluze image
Let d de, d da denola he
tour basieene
durectons for 3x3 Tegon

Dogmal i the ang

-67.sto - |2.S o +67-s° to 412¢ó

+ (S7 S to
or -g.sto -|2$°,
-157.5 +157.5° a) Too potble
Edge normal

3x> nughborhecd
P Pe

PPsP Edge
Edge normal
(b Rang Value b
(eradientvector)
K» the urecton
Se normal -22.59 +22.5° (b
angle oy the edge
-I57.5° +157.5
+45°edge
notmaI, torhnontal
-12.5° +112.5°
(cs The ange marngee o}
-Vertical
cdye Ihe diretonA
In a 3*3
-67.5° 467.5°
neghtod.
-45°cdge
-22.5 +22.5"

Horizuntal
cdge

ouble Conneutvig
threaht
deng
analyeito detect x dnk edqee'

Cannye alqostim hy bereia thruhtding


which Uses oo hreehotd
a?

a) oo threshold T
d
6)Hugh threahet T
The Oto the hegh to Lb hre abol ehould
2i'
too o hree to One .

0e usulie the threshol


can dung operaton
too odditonal
|CreatinY tne ee

hoye fecoornon emo pxele

beeaue it intor
med with
louou treaod
By tn9
Non o prelifrom qnH (i4) elinunolid

A4t threaholdunq operatone,at kérongpxela


aumed to be Valud eqe potoe
Ptxele mokod (mmeddtely

Bounday Deteetion-
Loal pro teung
Tuoo propurtiu
edge king'
un
the atrengs(magnitude)g tha deleeted
edqe pont
thut arettont
yaually
done e loca neghborhecda

* Adjaent edge ponz oith


imloTmagrude.
durecibn inked.

eam ple)an Co ordunol


JS JoY edge pxel oit5
pre depned neLghbor
hDOd

nagtwe
threuld
A à a no0
threold

% Boun dary Detetioo


Gobal procaesug vo tthe tough Trarforro

Hough trana4o wocy tendng edge


pontz th Qn thot ie
tmag along
Braugh dine.
Example V.s Qb plane Cparam et
Rpau)
a

The tough tanaom


Vawu o
pare
eqat on that pau though (7y
heae ane

oilt Cook
uohen plottedtheae pauia o4 k

(7y)tocutons

oIRulo
parametei2afion Snusodo uTves Drvieion acumulate
plane tito
io the 3e-plane
plane

foc 2traughtine,
on
the
Atthough the
tough Erana torn applicateto, any tuneion
whre
Ve ctor al Cordnl
Vector ey Co-estuanla
For on He icle
2

An m
-tough tranator
totlooe.
bineryedge itnage Uinig any
teehnue
Speuythe b duVIRrOne tn Bo- plae.
3 ELamube he Counta o; CLCumu laloxCelWa
for hqh pil Contontratonu
Era he gelateonship Contnil)
betoeen
A
mene

Chooen celt .

THRESHoLDING.
Single Thxeeheld
Consider he intenait Hu lberam
gure
A aumptibn: Tha Range a inencila levela Covered by
Obeeli % hterst
is
diterenttrom baekarouod.

No: Ay poi ()
nt
at
Caled am
T objettoint,
Othruoue the pnt u
Multple Thre shold Called
baukgroundpant
where are ant three dutnet thlenaiy
Vaue

Globel Threaholdg"
hen Conetant appicable OVer an entre
the globed
proceu
theahotdny
Naiable Threzholdung/Regunnalthresheldinalkocal
O4hreholding
hen the Vaue of Hhree hold T

Changes Over an
rmage it reyteredto
a VaraŠle cool/
Rugeoras thresholdins

Kole Netie in Image Thresholdung


Nout lee
8- bitirnog
(6) Imag
addettveGouan

Lo tntneleelo
(c
wil
Emage
IGuan
addutvie
Nehe uoi
SD So
means0,

ct)Coreapon
|dung hitorae

(3 gimple ythetoimage free

ahouse ie hatorom
Sugmening
he rinag thto too regone Lhvotva

placéng
Hreehold any cohere betcoean

Corupled by
shouo Oruginod mage
the Atanclarddwah
Zero meon
Gauuoin

shous tb Hotorae the


betcoeen
A +hreshetd plat
ed

peat coould Begmern tnag

SO

ch enaile lvela.

a kufoo
toitoutadditonalproeina fhding

thiaimeae in dulgut

Rote Jllumnton ctne


Ratle

b) TntensiLf
amp

(C produet
(au (b

Hotoamg
The image
htogra
win a Ange threehot & .
Atogaayyaste
Intnalyramp tn Ra monge [o2 o &],

hutogram
shooe
Fe (cshooa produet (a)eCb)
Hotoqram: The deep Volleybetuoeen the
16
peaka oDi Corupld to
addetonal pro tn
Ihe poit ohee e peration
Long

|poale.
Boic Globod Trreahol
when

poibleto
Ehe

back an0und pxele


dang

ainterai
are
URe
l
Bnqle
dútribukt
on af Objeca

A4ficentydutinct
gCobod
,it
& the

d appicable
threahel
Ovr the entre (maa e.
most appiatorn. there u
enouh Varobiluíbed wun imagee an algerithm
automatreally thehold ne
Copable esbnatng
Vate each tntage n

1teratYe algorithinlanbe Uaed for thiu


|:Select an nitial eetimoali(or.aobalthreuhed T.

Ucung T.thi uoil


produce
hee Image
too
Gi Conauting og ale pixel yoithvntenai
Vaw T.

Cmean) ibten
the avurage
3
Compul
Voee m & m

Yupetvely
4 Compule nes threshold value:
Cmitm2).
UntHl the duerene
Repat though t iteratHong
betweeo th BucceAelve
Smaler thon a predognel paramdi dT
to Contro! the umbe
AT Ueed
Parama
a

Oh teTattonaiohenre kpeed is Lmportant

chtenal the image


The averaqe
rbittalChoce T.
Regór Based Segmentation.
threshelda Lome-imee do not qve good

YehuG tor kegmentation


the
Kegoin -b aaed Segmentahon
Connectvilg &imlo pLxela a region.
Each must be
Unigorm
Connecthuily o the priela uoithLn the.regon

Vey inportant
There are iop maln approachee to
based egon
Segmen lathon;
i) Region groorng
Basic Formulatiton
.

ket epreeent the enttre


R

Seamentation procu thod- pOrttons


R inlo Subegona, Buch hat

ld) pCR:)=True i:t,2n


(e
PC RE UG) Falue for any adyaunt eagron
where PCRR3 a Logicad preclicaledeyincd
pona m Bet RK
Dver the
True ! Pxele tn Rk hare the
for ea PCRK) aLe

Bame level

Regon Grooing2

Reqon roolg poedure that


chto lorqenr megionabaged
pized 3b Tegcon
On ton
predu gined Criterei grooth.
to atart i(thn
Seed" pona and fom groo regone
theae
thoee nughborI
bt appending to Beed

plxele have pre datned propvHie


fo he Beed

Let denolt an input cmag at


'2
Beed Contaunin
S(* y denolê
4he locotion Seed pornTa
appued tot eah
9 denole a predieale to be

louation C9 ol Lame 2iBe


Osned to be

on
A boiu egeon gou0in algpthnboned
be
8- Cooneetly may foleuoas

Conneled Cornponent tn S(y)and


erode each eocsp Connecled component to one
Plxel:label au Buch plx found
othes pue S are Labe lled
Form Buch that at Pair
Lmage da
Co ercinalii
-4he Q, at
predicali,
geven
Iinput mag
those Co odinala,o4heruoué ket faliy2D.
3 Let be
image formedheby appending
I- Vaued
to Beed point
to thot
poina fe that are 8- Conneelid

potnt.
tn
|4 obel Conneld Componet

the
begmented imag obtauned by egon qoug

Reguón Spttng Merging


Lhe
Regon aputng
entre
Hhere in a
Lor vegon (posublethe
im age
(mosureni) UAed to
Then a predicli
determne eyon n Urnorm
hat the
not.then the method

hen eaeh o theue tuoo eqeoing in idepenaenty


tesledby predieati.
Thi pOtA dune Contmuu au
UntiIl raltng
5
The main poblern
delimningvohere to splita vegion.
eo to Uee
One metho d & to devide a eqcon

Quad e 3 Lee t
A bouch nod ea heve etac
our de en&an

R2

The kplit Mrge proedure


;
splitchto tour diajontquadro n ony
kifor which CRi )- fale
Yegcon
When {urtherspittng u poaible,mrg
ng adja unt eqcona Rj and R tor ohich

Q(RUR) True.
Stop ohen no further meqig or Rpluttny
poble

You might also like