Image Processing Unit 3 Part 1
Image Processing Unit 3 Part 1
K convoluttonal openatb
thu requeny domain, thL deqraded image Lgivon
Oiginal image
T p8sible b4 LmOvg
or munimizing degraclatimi
Type of noise
Noue s
atuuclualuon in pixel Values and
it characteüzed hy Yandlom yauable
umae
Jhe PDF of a Gjaussian landom vauable Z
guven by (z-uf/2.6
P
standard deviation
Vauanc
c co
foY
JoY a
a Gaussian nandom yaiable ,
the probabiliy
3 aYe in ange z t o L
that values of
app roxumately. 068
Zt 20 is ahout 0 95,
Impulse Noue
It u also
known as Shot noise., Salt and. Hppert
moise and Binay. noise
t
ocus mostl hecau of AensD and momo"
problen because. owhuh pixels ase Aslgned
incoeet mazimum. Values
for = 2-1
P= fo 7 o
-(Btp) fon z =V
inteae valuu in the ang
where V i l any,
oV 2K-.
und sudd en
This noise can be caused. by Shasp
in the tmage signal. t presenlt ilbel
dishus bances
whte and
black pizels
AS Aparsely vccuing.
3 Poisson Noise
t o t a l numbun pirelu
P atio o nolsy pizels to the totl mumb
4 Ezpon.entinl Noue
This
Jtpe o nois ouurs mostly due. o the
poblems t a ilwninatucon
prusnt lase unaguna
n
P)
o othenwLse .
Mean- Yiance>
5 Gamma Notse-/lang
Nouu
Ih
type onou aso occuss mostly dus to he
ilnminahion pro blms,
P Z 20
(b-9
Z0
and
6 Rayleigh Noue
-
Pz)
Pz)- -a)7 b Za atTb4
7 b(4-7T)
LUnlovm Nos >
Pz- a 4 z 4 b
othewis
Z tb_
2
nd b-)2
12
Type c hased, om Conelation
Statistical dependente
oYYe lati on
amona piels L Rnoun O
a
pixcl Ls ndependent ot t
t
Rnounas wrwelated. ptxel
neighboving piselu
othew
s Rown a s coYYelated. .
0 PlicKer Noise
It can he modellec as
multipucativc protes
Speckle noise ii most -enLounteud.
multipuiat1ve.
nose in
poe siinq
Tt is mosty pesent n medicaltmaqeA
Tcan he modelled as pixel yalue TABd multiplied
T-
ftuy)t Lft») xNg
Kanddom MouL havtng, a zero meangaussiann PDE,
Mean Hilte
1 Arcthmetic_mean filte->
emovs Local
yari uons within the ima4e
This ltey
(st) ES
-> masK Lapplted over bloret imaq.e
where g(St)
ís
Ex: foa3 X3 image the m asK
11L
9 1
3 qrom ebu m ean ilte
nolSe
This filt er elminates Gaussian
mn
fay)71a)
(E) Tmage 1 2
46
78 9
m=3,n=3 /9
(1x2x3X4X5X6 X¥X8x9
. )
(Ez)-Tnperttmage 152 36 fg)
8 99
4 Conbua -
harmonit. mean
Hter
t well auited. tor reducing the ech of Anlt
And peppet noise
2,gt)
(s,t)E.Sy
atthmetic meanfilter
when. 6=0 ,
it becomes
hamonic mean filter
,-1, i t becomes
ltns
Autrd for umpule
>Contrahasmnic
Yalue of eachpoxelL
the color
. mean fikor,
wnbahamoni
In
nilth color values oth
tontrahaimonlc
mean o
veplaed
in a Sumounding Yequom_
d
pLLel
01dey Statuáti. iltens->
fy)-(St)E
min(6,t)
S
4) Mtdpo int itteu>
+ min 1gt)}
m f ft) 2 mAx 1g(s,t)
(S,t) ES (St) ESu
m e a n filte
5) Alpha tilmmed.
It i baseed on the Concept o Computalion othe
that fall within t h u windour
AVeraa othe pixels
lo westandthe el2
Tt workL by cdeleting thu dh
lwel Vaues
tighest 2re
I iA useful in situations involuing Tmultipli type
Auch ClA the Combination of- Aatt a n d
omouL
Peppe and qausian Moise
1
d 2 , ?r(6,.t)
mn d
(s,t)E Szy
d-o auithmetic m eanfilte
Td-mn-1-> medianiltet
6 ummed, mean filter -
Lt woyRA by removing the may Qnd mun values and
caleulating the ayerage. of. Jemainunq Value
3 4 5 6 8
-(5)
uestun L-Assume that the image qiven belour i attakd
both noises
4D 49 Ovder 5 X4
4 1 8 8 m 5
1 166
o5
mn=5X4 = 20
1 15 6
8.8,9,
0,0.L,1,1,1,1,1, 4, 4,4, 5,5,6,6,6,6,
9values
values
t Apha tvimmed
fty)= S, mean lterc
Tnnd
s.t)E Sy
1 444 2
18)=4.
20-18
Units-4t
Adaptiye , local Noise Reduchon. Hte
tontiagtUn t a l ,
Jhe v a u n e quves a measwu dt Lmage
Notati on
2
On
f)= )
BackaroundMatha natus fo this
e t n denote a discrete nandom vayable.
n m plar)
wheve m-) i plui)
The mean il a measwuL o yeaqe. mtensily an
the vaLance a measw
a
o mage contrast
2
Hy (a-m plat)
We conAider luwo use o tha mean andl v.auancet
enhantemen puY poses
9iven Lmaqe
S
given by. the ezpressLOn
Psu(i)
similany ven hy
L
2
(-msu) s
D
(x
Adaptive Median flte
he
meclian tilter pen foims well it dhe.
4pataal. denauly
heAalt .and. pepper mose is low. (B and.lp less than "2)
he
adaptive median filtenin c a n handle. moise ith
The
Jhe adaptive median ltet also WoRL n a rectangular
meighborhood Sxs
The median filtte shange-(inceasea) the siZ
alaptive
lterun4 dependhg on certaincondition
S ing
single value usedto
The output of the i l t is. aa
TS ma YepeatlevelA
Else outpub 3med
qveater than 1.
be impulaLye.
fthe Con.dition Z,
min Zy <Zmax dtusth@n the
Pixel a t Zy cannotba ha intensily oan mpulse ars
not impulle
qawanteL that tHese Values
aie
Jhere L mo
J
value, t h
the alqovithm outpuk
a
tvery ctame
tenter nghhov hood Sy mo Ved to the naxt
tho neighbo1hood