UNIT-II
IMAGE RESTORATION
DeHrniion-
which we get
Restoxation 1s a process in
Image
thc degaaded image
mage
af 03:9inal image faorn
appxorimation
appxorimation
filters.
xCstovation
by using
Model:
Image Degadakion /Resoxation
Tmage
y) Degadotion|
g9 Rcstozation«
function
H
rtcY
image and 'H
19 03xiginal
I n above block diagzam f(xy) omPonent
a no; Sc
1S a degxadation function , no1)
s
and g,y) 1s degadcd image by cnvoing
Gnvoving
inage g y ) iS
obtained by
The degsadation Com Poncnt
rotse Cam ponent
funckion H' along wiEh
degsadation
y) to the oginal im age f(XY).
+n(N9)
99) fA9) * hcxy)
F(uv) A (u,v) +N(uV)
G7 Cuv)
19 c p r e entad s tn focg domain
foe2. domain
The degsaded tmagc
wortten in above.
lhat 1s
domain
1S in facg.
whexe F(uv) The îmage
Sscg. dorrain
HCuw) = degaadaion n în
în facg.dornain
omporrent
N u v ) = Notse
Cnit-3, 13
fry) denofcs appaoximation
Tn he above blook diaqsam
that is obtainccd by applying
o he 0xiginal image
xcstosation
/p to
degsade image g(Yy) as a
Alhebxaic Resfovation'
Rcsfodadion 15 of tuo ypes.
Algebraic
) Censloaincd Reskoxation
Onconsfaincd Restoxation
Conskxained Restoxalion:-
assume
CXsor ) Let us
(Mean Squaste
Minimi3c
15 I1g-H?
The Mean squane ev0y
Le as assurnd C? a#l+ l9-HI
Thc Constöair
The atove expression equatino to
7s Zero.
Reskovation o funcion I n that we Should substac
the nois Cornponent
Can
let us assume a Conskant then the above
be wtte i as,
To Mintmie the emor We have to
to Con5idc
above
0Xde desivative and
Cquating t to the zexo.
nit-3 9l13
) la?1 :5 fn
a of n-Ig-HI
d;
laf+4 il9-H#-I}l=o
af-2« (H) (3-H?> =o
26 2kH (3-H) =o
aQf- 4gH 4 4HH =o
laa f HHTË]- «gH"
f aa Y+HH]-9#
e us assume =1
9H
Cac HH)
[aaTy+ HH] 9HT
f
Unit-3 313
The above a funcbion cn 15 the inal op of
Corns brained Rc storalion
Constaained
Estimaled ofp s apphed by usingy
Rcsforation
On -ConsBaincd Restovation'-
we know that fom thc basit block diaqam
of deqadakion Moclal
g-HE
Restoration 1s used fo Mintmiie
An un-ConsBaained
Hcan S2uaNe e1NO
( Inl - Ig- Hf*
nnT lg-HfI (g-Hf)
to zso by t
By eyuatinq mcan 5guane es1N0a
0dder dcsivative.
19-'] -o
-
2H (g-H? ) =0
A
-9HT HHT -a
HH HT
Unit-3 4l 13
InveYse FiHcsing
sestor
*This v e s e illexing, 1s best approach to get a
estoxation,
op in his e an use uncon stsained
dcstosation cstoxe
accoding to the unconskaained
olp 15 epresentcd bby,
gH
expyession in the
the rez
4IC?.
eprescnt lhe above
let us
domain
GICav)
u,v) H (u,v)
FCuv) H(u) +N(u)
H(uv)
NCuv
tav) F(uv) + Huv)
NuN)
Fuv) Fuv Huv)
N (uv)
F (uw) Gluv)
E (uN HUV) H(u)
Oni-3 5h3
Cuv)
G7(uv)
f Cu v)
H(u v)
above Cn. we a1C no
Ho
* applying IFT
(y) oxnginal imoge
9elting eact
invexse Fouie Ezanstomn
CC fX,y)
*Because in the above cxp3ession we knoc
only
Cn
Component value s
degradation value and reise
mknocon.
* In the above cgnsF (uv)' denotes scq. dormain
epresentation of 0t9inal mage tx) and fse.
domain epresentation of degsadation image GCx9)
G (a 9)
and HCuv) is denotes ansfer turctio n [feg.
domain vepresentation of degaadation n 'H'.
Minimum Mean 5guae eror THCinGY] fltecx-
a Mcan S9uaxe
*weiner filkex 1s uScd fox Minimiaing
fn and he No'Se Compo-
Crros In this degsadation
as Random vaiables
ncnts ate Considcscd
Thercfore the cstore op s obtaired by Ninimiäina
the Hean Suaie erod S
Mean Sguane croY
epyesentad by
where f undeqsadcd image
Restored output
Unit-3, G13
Restove o/p cp»cscntcd m regucncy dornain ith
undegvaded
he hep of fowcx specl>al densily of
of NoiSe
fmage and power SPccloal densy
fcu) Ga Cu)
IHN)l's uw)+S,w)
(uV HuNw)
Snu,v)Gr(u GtaN))
°FUN) (HtuND|+
HCuw= H(uw) H*(u N)
IHOu M) Cn(uv)
fauv) H(av) Huv)+na
LS (uv)
xanform degade
dqade
denotes
the above cxprcssion H(uv)
Tn
function to 'H
undegaaded mage.
mage.
dcnotes PsD o
Cu)
S omponent
Notse
denokes PsD of fransfam of
of
dojadation
degraation
Sn (4v) Coniugate of
Conjugate
xansfam
denotcs Cornple
Hu»v)
&cduce
cduce the noise
noise |
we
wC Can
Caan
f'n to no:se
otio
Using 5ignal
Byy tc as
it as
SNR 5
i-e,
ComPonenf
IFawl
SNRMN N (a)
N
M
Onit 3, |13
nluv
whcxc 61 (u ) denotCs degradcd nage
o1Comcfdical Mcan ilfcr
The gcomelaical MCan fllcr is having ta 59uate
ilh a powCy d and1-<'.
brackels is having
lisalion af teinc f:lt c
a 9cncaa
*The xsBoxcd olp is eprcsenlcd b PCuv)
(u»v)
-
HuN)7* H (uN)
Gi(uv)
PCuv) H(uv IHCuv)+p Snuv
and cal Conslants
whec, B a1e +ve
diffexent valucs of we gek nuerse
*BB substituting
alkesstandaxd
mverse
hwe&se
WCinc&
Sller and Pasameter
faller and axo We iner hlHcr and by Subsliluting 'a'
'p'l then etse
valuc 1s greatcr han 2 («>%) and
gef geomckaical Mean filkcx olp.
above eg'n.we get invcase
1 in
) By substitulingd
1)
frltex op i-e
HAv)
Huv)1
ftuv) THuv*HCuv)4p [Sn(uv) G(u
HCu)
IHtuv) G(uV)
Tnvcse filkcs o/p.
Unit-3
) NOCw ubstitute d %. hen WC gct Po ramc taic
weiney tillcs o/r.
uv Huv)
Fcu v) -
GiCuv)
IH(u Huv)+B SSn(uN
[Sg (uJJ
Faxarnefxic weine illcr olp
tii) n oeder to gekting slandaxd einey frltcs olpP we
shoukd Substitute the valuc of s lesslhan %
(o ic, o) and B alue =
t-O
HCuN)7°r Hu)
fca Gu)
THCUND
IHuwLsuvJ
H Cuv)
uv) Hul Sptuw Gatuv)
P .
then
F l:l thcn
and
and
substituking the value of d
By
Mean hhcr olp.
itcs olp
9eometaic
we get
Mean s9uane /tcaing-
Corstaainad least
2uaxe iteatng is Mainly
least Mean
This on stoained that
that 15
15 one
one
iS
is Vasiance
Vaiance
Pasameters
on o
de pends
one isS Mean.
Second
and the block
Restoxation
to the degxadation /by
ACcoding
degadcd image s obtaine
know at a
we
diagxam .
(undegoaded mage) sth
with
cornvolving fx»y)
by
by
and noise Com Ponet nr y).
degaada tion f
unit-313
fo Convolvcd o
is addcd
9() F0y)* H(X»y) +nxy)
fom
vecor fon
Vectox
be wrile
witc t 09
0s
Can
expxession
7he above
Hf4 n>0)
whexc, H degradalion furdion
function
f Lmdegaded mage
n'
omPonent
n no}se
vation
must be
be Constoainecd
lonsteaine
e he csto
*To Meantnqtull
blems af
af hand.
hand Thus,
of the pxo
by the Paxametcrs Cxtasian
aitasian
find he Minimum
whaf 1s desired 1s do
defincd as,
kretion that 1s
H-1N-
Caitaian fn
where, C Min,mum
(ov) lapaBion opevator
v ay)= laplacjan
Constrained
Subiect to the
19-H= lIni0)
The Restoration o n fseg. domain 7s dePacsented b
fca)
unit-3, l6|13
H Cuv)
fcuv CnCa v)
IH(uv +x(puv)
1s YCPCCrled
7he mvexc of Power spcctxal densily
with Pr
PXy) -
)
-I
FSom he egn au) we Can Corncludc that b
difcsent
Ssubséituting diffeYcnt valucs of (y we
w e gek
TnVCase
nvcase
then if is egual to
Restoxation o P s I Xo
tle ing ofp. we should
to gelting belter Restored
*In oxde
define a esidual" as
Y9 -Hf
pa)=T
wherc '1s an Honotonically
.txt =lIn/l t a
csidual f'n
*T1s not a d:fficult to ind out a
bcttcr 3esfoxe o/P
of an image nosde
to getking
bc6ing he follocoing PYOceduae skeps we Can
Loni-3t13
csidual fn value
go tihe
)subsf:tude the value
) Conpute he residuc (lix1)
Pocess
when Ilil-ni 2a
lxll-ttn2a
Processer (ox)
iStop the
Continue the
the by
b
process
Sabisficdofherwise
i5
15 Safisfied, IlvnE tnh a
decreasing
'value if
il19 Sakisfied
' value
(0X) by tncrcasinghe
the con dikion I = IIn/t-a
value
5 zc0 lo)
Co)
whee, a' 1s Cornstan&
then we get
above shauld find out he values o
For the we
Ili and Iml
esiducin fe9. domain
*we kmow that the
RCuv (uv) Hu vof(u)
discacte fouic r txansfom to
By aPPlying invese
above cxpaession we get the xesidue in SPatial ox
SPacial domain.
Uni-3, 12/13
Jn above Suma tion ndiCafcs
CxpcsSion, double
double Hagnilude
subsl;(utde the
the
fox gelting desivad olp we stoud
of Cornponcnt
vakics of Vaiance and Mcan roisec
2 2n(y) -mn(y)
mn 2210y)
mnMN xo
CornPonent
where mn Smple Mean value o nise
Conclude that the
hat
Can
*FXom the beloc exprcssion w
Mean and
and
obtained by &abstitutinq,
optimized o is
vaiance of no1se.
unit-3, 1313
