IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
Imag Resto
Image
estoration
ration
Naku
kul Kumar Gu
Gupta
PG Scholar,
lar, Department
nt of CS & IT
JAIN
IN ( Deemed-To
To-Be-Univers
versity) Bangalor
lore, India
jpc2226
22672@jainuniv
niversity.ac.in
Dr.S.K.
S.K.Manju Barga
rgavi
Professor, Dep
Department off CS & IT
JAIN (Deemed-To
To-Be-University)
sity) Bangalore,
re, India
b.manju@
ju@jainuniversi
versity.ac.in
Publication
on History
Manuscript
ipt Reference
e No:
N IJIRIS/RS/
/RS/Vol.10/Issue0
e02/FBIS10080
Research Art
rticle | Open Access
A
| Doub
uble-Blind Peer
eer-Reviewed | A
Article ID: IJIRI
JIRIS/RS/Vol.10/I
0/Issue02/FBIS110080
Received: 244, January 2024
4|Revised:01,, February 202
024 | Accepted:
ed: 08, February
ary 2024 Publish
lished Online: 299, February 2024
20
Volume 20244 | Article ID
D FBIS10080
http://www.ij
w.ijiris.com/volum
lumes/Vol10/iss
/iss-02/01.FBIS10
10080.pdf
Article Cita
itation:Nakul,M
ul,Manju(2024
4).Image Resto
Restoration. Inter
ternational Jour
urnal of Innova
vative Research
rch in Informat
ation
Security (IJIRI
JIRIS),Vol.10,Issu
ssue 02,42-50
doi: https://d
://doi.org/10.265
6562/ijiris.2023.
23.v1002.01
BibTex key
ey: Manju@20
2024Image
Copyright:
ht: ©2024 This
is is an open acces
access articlee distributed
d
un
under the term
erms of the Crea
reative Commo
mons
Attribution
ion License; wh
which Permits
its unrestricted
ed use, distribu
ribution, and reproduction
rep
n in
i any mediu
edium,
provided the
th original author
aut
and source
urce are credited.
credited
Abstract: Im
Image restorat
ration is an integ
tegral compone
onent of compu
puter vision tha
that tries to restore
res
pictures
res that have been
be
deteriorated
ed or corrupted
ted to their orig
riginal or enhan
anced condition
ion. In this study
udy, we look into
int the wide-ra
ranging terrain
ain of
picture resto
restoration techniq
iques, whichh in
includes both
th conventional
nal filter-based
ed approaches
a
and
an cutting-edg
edge deep learn
rning
models. Ther
There are certain
ain circumstances
nces in which tra
traditional app
pproaches, such
uch as Wiener
er filtering
f
and
d bbilateral filteri
ering,
perform quite
uite well, particu
ticularly when itt comes to sm
smoothing and
nd noise reducti
ction. On thee other
o
hand,, th
the fact that they
th
rely on hand
ndcrafted filters
ers restricts the
their adaptation
ion to more co
complicated form
forms of degrad
radation. Visual
al restoration
n has
h
been revolut
lutionized byy deep
de learning,
g, which is led
ed by convolutio
utional neurall nnetworks (CNNs).
(CN
Deep lea
learning involves
olves
learning soph
phisticated representations
repres
so
of visual data.
ta. It is because
se of this thatt C
CNNs are able
abl to deal with a wide variety
variet
of degradatio
tions, such ass noise,
n
blurrin
rring, artifacts,, aand missing data.
d
Generati
erative adversaria
rial networks,
ks, often
o
known
wn as
GANs, are
re co
continually pushing
pu
the limits
lim of whatt is possible by utilizing advers
dversarial trainin
ining to accomp
mplish spectacu
cular
outcomes in the areas off in painting and picture super
uper-resolution.
on. Despite ama
mazing developm
pment, theree are still obstacl
tacles
to overcome:
me: Understand
nding the inner
ner workings of deep learnin
rning models continues
co
to be
b a challenge,
ge, thanks to
o the
t
limited interp
erpretability of the data. Dep
ependence onn data: Acquiri
iring large quan
uantities of high
igh-quality data
ta is necessaryy ffor
the trainingg o
of successful models.
m
Costs
sts associated wi
with computati
tation: The proc
rocess of training
ing and deployin
ying deep learn
rning
models may
ay be quite computationally
com
lly rigorous. The improveme
vement of camera
era vision forr autonomous
au
s ca
cars in order
er to
make navigat
gation safer and more depe
ependable overa
verall. Image res
restoration tech
echnology hass the
t potential
ial to continue
ue to
revolutionize
ize image proces
cessing and anal
nalysis, ultimatel
tely contributin
ting to advancem
cements across
ss a wide range
ge of scientificc aand
technological
cal domains. This
Th can be accomplished
a
ed by addressin
sing the challen
llenges that are currentlyy bbeing faced and
a
concentrating
ting on the prom
romising research
rch directionss tthat are curren
rrently being purs
pursued.
Keywords: Image Restora
oration, Artificia
icial Intelligence
ce (AI), noisyy data,
da grainy pho
hoto, Local noi
oise reduction
INTRODUC
DUCTION
The practice
ice of restoring
ing a damaged
ed or
o corrupted
ed image to as near its orig
riginal state as feasible is kn
known as imag
age
restoration.
n. T
This can bee achieved
ach
by reducing
red
the ima
image's noise,, blur,
bl artefacts,
cts, and other distortions.
dis
Rand
Random variation
ions
in an image's
e's colour or intensity
int
are referre
referred to as no
noise. Numero
erous things, including
incl
the camera
cam
sensor's
r's sensitivity,, the
th
amount of lig
light present,, and
an the picture
ure signal's trans
ransit, might cont
contribute to it.
t. Blur
Bl is the loss
los of an image'
ge's sharpnesss or
o
clarity. Numer
merous things,
gs, including the out-of-focus
cus lens, motion
tion from thee camera
ca
or subject,
sub
and diffraction,
diff
migh
ight
contributee to it. In digital
al image proces
cessing, imagee res
restoration is a crucial step
ep that attempts
pts to restoree a clear, origina
ginal
image from o
one that hass been
b
damaged
ged or corrupte
pted. It entails
ls locating
l
and
d el
eliminating several
severa flaws or degradation
ons,
like noise, blu
blur, artefacts,
s, and
a distortion
ions, that have
ve an impact on an image's quality.
qu
In severa
everal fields, incl
ncluding as digita
tal
forensics, sat
satellite imagee processing,
p
, as
astronomicall im
image process
cessing, and med
edical imaging,
g, restorationn ttechniques are
essential. Tra
Traditional imag
age restoratio
tion methodss o
often rely on mathematica
tical models and
an statistical
cal techniquess to
t
estimate the
he underlyingg original
o
image.
age. Despitee th
their remarkab
rkable progress,
ress, deep learnin
ning-based imag
mage restoratio
tion
methods still
till face challeng
nges, such as the
th need forr la
large amounts
ts of training da
data, the riskk of
o overfitting,
g, and
a the lack
ck o
of
interpretabilit
bility in the learn
earned models.
s. FFuture research
earch directions
ns include addres
dressing thesee challenges,
ch
developing
devel
more
ore
efficient and
nd robust deep
eep learning arch
rchitectures, an
and exploringg tthe applicatio
tion of deep learning
lea
to other
oth challengin
ging
image restora
toration tasks.
ks. Image
I
restora
oration is an evo
evolving field
eld with significa
icant implication
ions for variou
ous domains.. As
A
research pro
rogresses, deep
eep learning is po
poised to play
ay an increasingl
ngly crucial role
le in enhancing
cing image quality
lity and unlockin
king
new possibilit
ilities in imagee processing
p
and restoration.
on.
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 42
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
A wide range
ge of methods
ds are used in th
the vast subject
ject of imagee reconstruction
reco
n to improve
ve or
o restore pict
ictures that have
ave
been damaged
aged by noise,
e, blur, or other
ther artifacts.. A
Applicationss ffor these methods
me
are numerous
nu
and
nd include imag
age
processing,, aastronomy, and
an medical imaging.
im
Iterat
erative reconstru
truction is one
ne of the most
ost widely used
sed methods for
fo
reconstructin
cting images. An initial estima
imate of the pic
picture is used
ed to create a nnew estimate
te in iterative
ve rreconstruction
ction,
which is subs
bsequently used
sed to createe an even better
ter estimate, and so on. Until
ntil the estimate
ate approaches
es a solution tha
that
closely resem
embles the origi
riginal image, this
thi procedure
re iis repeated.
Generalizab
zability: Models
els that have bbeen trained
ed o
on certain datasets
dat
mightt not
n perform
rm well
wel on data
ta that they have
hav
not previousl
usly encountered
ered.
The followin
wing are some
me potential
al ffuture direct
ections for research:
res
To understand
u
the behavior of models and
d tto
develop conf
confidence, more
ore interpretabl
able models are needed. To lessen the
he dependency
cy on massive
ve datasets, data
ataefficient appro
proaches are
re being
be developed
ped. In order
er to reduce the
he amount off computing
co
resources
reso
required
uired, lightweigh
ight
models are
re bbeing developed
ped. In order
er to improve
ve gen
generalizability
ty o
over a widee variety
va
of degra
egradation types
ypes and domain
ains,
domain-agnos
nostic models are being devel
eveloped.
As picture
e rrestoration
on technology
gy continuess tto advance,
e, iit will enabl
able larger use
ses in a variet
iety of sectors
ors,
including th
the following
ing: Improved
ved diagnosis
d
accu
ccuracy and the ability to vi
visualize essen
ential details are the goals
ls o
of
medical imag
aging. The restoration
res
off satellite
s
and
d aerial photo
tographs forr th
the purposee of
o improving
ng land use an
and
environmenta
ental monitoring
ing is what is known
kn
as remo
remote sensing. Another
An
significa
ificant method
d for reconstru
tructing images
es is
compressed
ed ssensing. A smaller
sm
number
er of measurem
rements are used
sed to compres
ress the picture
re in compressed
essed sensing tha
than
what is nec
ecessary to accurately
accu
capture
cap
the im
image. A scari
carifying transfo
sform, which converts
co
a ppicture into
o a
representatio
tion with few non-zero
n
coef
coefficients, is ap
applied to the
he image in ord
rder to do this.
this An approx
oximation of th
the
original image
age may then
n be
b constructed
cted using the compressed
ed data.
d
A relativ
relatively new kind of picture
re reconstructio
reco
ction
method that
at has shown promise
pro
recent
ently is called gen
generative advers
dversarial netwo
works or GANs
Ns for short.. With
Wi insufficien
cient
or poor qua
uality input, GANs
GA
may pro
produce high-qu
quality pictures
res. Image reconstruction
reco
is one of thee most effective
ctive
methods inn co
computer vision
visio for fixingg d
damaged orr d
distorted pictu
ictures. And for generating new
n
images from
fro incomplet
lete
or low-quality
ality data.
These tech
chniques have
ave a wide range
ran
of pote
otential applic
lications, Inclu
cluding: Image
age reconstruct
ction is a usefu
seful
technique in computer
er vision
vis
that may
ay be used to create new
ew pictures from
rom missing orr low-qualityy d
data and repa
pair
damaged orr corrupted images.
im
There
ere are severall ppossible uses
es for these met
methods, such
ch as:
a Medicall imaging:
im
X-rays
ays,
MRIs, and CT scans may
ay all be improved
roved by picture
ture reconstruct
uction. This can facilitate better
be
diagnosis
sis accuracy an
and
make it simp
impler for med
edical professio
sionals to mon
onitor the cou
course of illness
esses. Astronom
omy: Degraded
ed astronomica
ical
photographs
hs due to noise,
ise, blurring, and other artifa
rtifacts’ can bee res
restored viaa image
im
reconst
nstruction. This can aid in th
the
study of far--off objects and
an the discovery
covery of new
ew o
ones by astron
ronomers. Image
age processing:
ng: Reconstructi
ction of pictures
ures
may be utiliz
tilized to impro
prove and work with photo
otos for a vari
ariety of purpo
rposes, such as artistic exp
xpression, imag
age
compression,
on, and security
ity applications.
LI
LITERATURE
RE REVIEW
The purpose
se of this review
revi
of thee lit
literature is to give a thoro
horough overvi
verview of thee area of image
ge reconstructi
ction,
emphasizingg tthe many applications
app
and
nd significant cu
current research
earch directions.
ns. Rather than
an emphasizing
ing well-establish
lished
basics, it high
ighlights novel
el discoveries
d
and recent advan
vancements.
Reconstruc
uction through
ugh Iteration:
Deep Learnin
rning (DL) algori
orithms have
ve bee
been explored
red rec
recently forr it
iterative reconstruction,
recon
with
wit the goal of achieving better
bet
performance.
ce. Research looks
loo at usingg pre-trained
p
D
DL models for image priors
rs in compress
ressed sensing MRI and examin
mines
how well recu
recurrent neura
ral networks (RN
(RNNs) functio
ction for PET
T reco
reconstruction
on that is iterati
ative. [1] Furtherm
rthermore, there
ere is
growing inter
terest in adapti
ptive regulariza
ization techniqu
iques for iterati
erative CT recon
reconstruction based
bas on learnt
nt dictionaries
es for
noise reducti
ction.
Figure:
igure: field and
d aapplication of iimage reconst
nstruction
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 43
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
A key compo
ponent of digit
igital image proces
rocessing is ima
mage restoratio
tion, which attem
ttempts to recreate
recrea an origin
riginal image from
rom a
damaged orr co
corrupted one.
one This proced
cedure entailss lo
locating and removing
rem
a vari
variety of flaws
ws or
o degradation
ions, such as noise,
no
blur, distortio
rtions, and artef
rtefacts, that imp
impair the image
age quality. In several
severa fields,
s, including
i
ass digital
d
forensics
sics, satellite image
im
processing,, as
astronomicall image
im
process
cessing, and medi
edical imaging,, image
im
restorat
ration techniques
ques are essenti
ential. Conventio
tional
Techniques
es ffor Image Restoration
Res
in order
o
to appro
proximate thee underlying or
original image,
e, traditionall im
image restorat
ration
approaches
es m
mostly rely on statistical tec
techniques and
dm
mathematical
cal models.
These techn
chniques can
n be
b divided roughly
ro
into
o tthree prima
mary categorie
ries:
Filtering-based
ased techniques
es: These meth
ethods use differen
ifferent filters
ers to
t take outt tthe image's blur
bl and noise.
ise. Wavelet-bas
based
filters, adaptiv
ptive filters, and
nd linear filters
ers are examples
les of common
on filters. Wherea
ereas adaptive
ve filters change
ge their weights
hts in
response to
o llocal imagee attributes,
att
linea
ear filters alter
er the image by applying a wei
weighted averag
verage of nearbyy ppixels.[2]
Model-base
sed technique
ues: These tec
echniques recon
reconstruct thee o
original image
ge by using pree
reexisting knowl
owledge about
ut the
image structu
cture or thee degradation
d
pro
process. Invers
erse filtering,, m
maximum a pposteriori (MAP)
(MA estimatio
tion, and Bayes
yesian
inference are
re a few examp
mples.[3]
Optimizatio
tion-based techniques:
tec
U
Using
thesee ap
approaches, the restoration
on problem iss formulated
fo
as an optimizat
zation
problem, and
nd the target
et image
im
is found
nd by minimizin
izing a particular
lar cost function
tion. Geneticc algorithms,
alg
simu
imulated anneali
ealing,
and gradient
ent descent are
re examples
e
off co
common optim
timization techn
echniques.[4]
Deep Learn
arning-Integra
rated Image
e Restoration
ion: When it co
comes to hand
andling complex
lex degradation
on scenarios, deep
learning has
as proven to be
b a more
re effective
e
tool
ol for imagee restoration
res
than
th tradition
onal methodss iin recent years.
yea
Convolutiona
nal neural netwo
etworks (CNN
Ns) are used
ed iin deep learn
earning-based techniques
tec
to analyze
a
and res
restore damag
aged
images by dis
discovering their
eir underlyingg ppatterns and fea
features. The key advantages
ages of deep learning-based
lea
im
image restorat
ration
methods incl
include: End-to
to-end learning
ing: Withoutt re
requiring ma
manual feature
re extraction
n or
o prior knowledge
kno
off the
degradationn pprocess, deep
eep learning mod
odels can bee tra
trained directly
rectly from degrad
raded images and
an their corres
rresponding grou
round
truth clean im
images.[5]
Representa
tation learnin
ing: With their
eir ability to lear
learn intricatee representations
rep
ons of images that
th encompass
ass both high-lev
level
and low-leve
evel features,, deep
d
neurall nnetworks are able to ha
handle a great
reater variety of degradation
ions and produ
duce
restorations
ns tthat are more
ore precise.
Generalizat
zation capabili
bilities: Deep
eep learning
l
model
odels are capabl
able of perform
rming well on
n a range of im
image restorati
ration
tasks because
use of their stro
trong ability to generalizee to new data. As we are plan
lanning to study
dy the noisee in our image wi
with
the help off aadditive gauss
ssian noise tech
echnique We w
will first use [6]
[ synthetic
ic additive Gaus
aussian noise to investigatee the
t
impact of corr
corrupted target
rgets. We emplo
loy the L2 loss
ss for training in order to recover
reco
the mea
ean becausee the noise has zero
zer
mean. Denois
oising performa
mance (dB in KO
KODAK dataset
aset) as a functio
ction of training
ng epoch for additive
add
Gaussia
sian noise.
Figure1 : def
deferent typee of noise in imag
age
Figure 2.1: ( Caucasian)) The
Th convergence
ence speed and
d final qualityy of
o targets with Gaussian noi
oise,
Clean
ean targets, and
d noisy targets
ets are fairly com
omparable.
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 44
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
Figure 2.2: We find that for
fo brown Gau
aussian noise,
e, th
the convergence
ence is slowed
wed d
down but the
he final performa
rmance is nearr due
d
to great
reater inter-pixel
el noise correla
rrelation (wider
er sp
spatial blur; on
one graph per
er bandwidth).
b
Figure 2.3: Impact
Im
of varyin
rying a fixed capt
capture budgett allocation
al
to cl
clean vs. noisy
isy cases (see
ee text).
tex
Poisson noi
oise: is the main
mai cause of no
noise in pictures
ures. Since it is d
dependent on the signal,, it is more difficu
ifficult to elimina
inate
while zero-m
mean. Through
ughout training,
g, we adjust th
the noise magn
agnitude λ ∈ [0, 50] and employ
emp
the L2
2 loss.
lo Training
ng on
clean targets
ts yields a good
od result of 30
30.59 ± 0.02 dB, while trainin
ining on noisyy ta
targets also yields
yiel a good
d res
result of 30.57
.57 ±
0.02 dB at a similar converg
nvergence speed
eed. Binomiall no
noise, also kno
known as multip
ltiplicative Bernoulli
Berno
noise, crea
creates a rando
ndom
mask m, wher
where 0 represen
resents zeroed or missing pixel
xels and 1 repres
resents valid pix
pixels. As expla
plained by Ulyan
yanov et al. (2017)
(201
in the contex
text of their deep
dee image prio
rior (DIP), we eeliminate miss
issing pixels fro
from the loss:: argmin
arg
Xi (m
m (fθ(ˆxi) − yˆi))2,
yˆi))2
(7) in order
er tto prevent back propagatin
ting gradients fro
from such pixel
xels.
Figure 3: Removing
Rem
Poisso
sson noise from
m image usingg wavelet
wa
doma
main
Text remov
moval: The corru
rruption is mad
ade up of many
any, randomlyy aarranged rando
ndom strings that
th are stacked
cked on top off one
o
another, in ra
random locatio
cations, and with
ith random fon
ont size and color
co
variations
ns. Both the font
fo and thee st
string orientati
tation
don't change.
ge. high dynami
mic range Thee floating-point
f
int pixel luminan
nance can vary
ry by several orders
ord
of magn
gnitude even wi
with
sufficient sam
ampling.. To produce
pro
an ima
mage suitablee fo
for the usually
lly 8-bit display
ay devices, this
is large dynamic
mic range must
st be
reduced to a set range using
usin a tone map
apping operato
tor.
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 45
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
Figure 4 : ex
example of the
he text on imag
age
Denoising
gM
Monte Carlo
rlo rendered
dp
pictures: We trained a denoiser
den
withh 664 samples per pixel (spp)
p) on Monte Ca
Carlo
path traced
ced images. For
or validation,, w
we chose 34 photograph
phs from a sep
separate collec
llection of situa
ituations, and 8860
architectural
ral images made
de up our traini
ining set. Three
ree versions off tthe training pi
pictures were generated:
g
on
one with 131kk spp
s
(clean target)
et), two with 64 spp (noisyy ta
target, noisyy in
input). These were made wit
with various random
ran
seeds.[7
[7]
Figure 5: Conversion
ersion of the blurr aand noisy ima
mage into clear
ear aand recogniza
izable image
Figure
ure 6 : For rando
ndom impulse noise,
n
the appro
pprox. mode-see
seeking L0 loss
ss pperforms bett
etter than thee m
mean (L2)
or med
edian (L1) seeki
eeking losses.
Monte Carlo
rlo Rendering:
g: The
Th most com
common metho
ethod for creatin
ting physicallyy accurate
a
virtu
rtual environme
ment renderings
rings is
called Monte
nte Carlo path
th tracing. To do
d this, onee m
must createe arbitrary
a
"ligh
"light paths" in
n the
t scene by connectingg light
li
sources and
d virtual sensors
sors, then integ
tegrate the radia
radiance that each path carries
rries (Veach & Guibas,
G
1995).
5). Because of the
way the Mon
onte Carlo integ
ntegrator is buil
uilt, the samplin
pling noise is zero
zero-mean and
nd each pixel's
's intensity
i
is determined
de
byy the
random path
th sampling pro
rocess. Decades
des of studyy int
into importance
nce sampling me
methods have
ve yielded
yi
littlee m
more insightt in
into
the distributi
ution, though.
h. It can be arb
arbitrarily multi
ultimodal, varies
ies from pixel
el to pixel, and
nd is highly dep
dependent on
n the
rendering pa
parameters and
an scene configuration
con
Co
Concentrated caustics and
nd other lightin
hting effects ca
can also produ
roduce
extraordinari
arily long-tailed
ed distributions
ns with sporadic,
dic, brilliant out
utliers. [8]
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 46
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
Figure 7 : Processes
es of constructi
ction of imagee fro
from signal
Magnetic Res
Resonance Imagi
aging (MRI): Fou
ourier transform
ormation of the
he signal, or its "k-space." Modern
Mo
MRI met
methods havee lo
long
employed co
compressed sensing
sen
(CS) to get around
d tthe Nyquist-SShannon restri
restriction. To be more precis
ecise, CS is used
ed to
under sample
ple k-space and
an perform non-linear
no
reco
reconstruction,
n, which uses
es the
t sparsityy of
o the picture
ure in the prop
roper
transform do
domain to reduce
redu aliasing (Lu
(Lustig et al., 20
2008).It is observed
served that the
he fundamental
al idea
i
remains
ns unchanged
u
wh
when
the k-space
ce ssampling is transformed
tra
int
into a stochastic
stic process with a certain pro
probability den
ensity p(k) over
ver the frequenci
encies
k. The k-spa
pace samplingg process
p
is sp
specifically descri
escribed as a Bern
Bernoulli proces
rocess, where
ere the
t probabilit
ility of choosin
sing a
frequency for acquisition
n is
i p(k) = e−λλ|k| for every frequency.
cy. Retained
Re
frequ
requencies are
re weighted
wei
byy the
th inverse of the
selection pro
robability, wherea
hereas non-selec
lected frequenci
ncies are set to zero. This "Ru
"Russian roulet
lette" strategyy is obviously what
wh
is anticipated
ted. The valuee λ determines
nes the overall
verall fraction off kk-space preserved
reserved. In contrast
cont
to a co
complete Nyqu
quistShannon sam
ampling, we set the valuee in the experim
eriments that follow
foll
to retain
ain 10% of the
he samples. The under samp
mpled
spectra are
re tra
transformed to the primall ppicture domain
ain using thee sta
standard invers
verse Fourier tran
ransform.
FFigure 8 : an illustration
ill
off aan input/target
rget image thatt is under sampl
pled, together
er with the matc
atching
fully sampled
led reference
ce an
and their spect
ectra.
We do ourr in
investigations
ns using 2D slices
lices from thee IX
IXI brain scan
can MRI dataset.
et.5. To simulat
late spectrum sampling,
sa
wee ppick
random samp
mples from the
he FFT of thee (a
(already recons
reconstructed) imag
ages in the dat
ataset. Consequ
equently, ourr data
da has a genu
enuine
value and incl
includes Thee periodicity
peri
off the
t discretee FFFT is differen
erent from thatt of
o real MRI
RI data.
d
The train
raining set includ
luded
5000 256x25
256 quality phot
hotos from 500 in
individuals, wh
whereas the validation
val
set inc
included 1000
0 randomly
ra
chos
hosen photos from
fro
10 differentt pparticipants.. The
T baselinee PSNR
P
was 20.0
0.03 dB whenn tthe poorly sam
sampled inputt images
im
were
ere di
directly recrea
created
using IFFT.
T. Th
The network
rk required
re
300
0 epochs
ep
to trai
rain. The averag
verage dB obtained
ned by the netw
etwork trained wi
with clean targ
argets
was 31.77,, wh
whereas thee network
net
trained
ned with noisyy ttargets had an PSNR of 31.7
1.74 dB on thee validation dat
ata.
Figure
re 9: Image reco
reconstruction m
model of reduci
ducing noise
This training
ng with clean targets
targ is compa
parable to previ
revious research
rch conducted
ed bby Wang et al.
al (2016) and
d Lee
L et al. (201
2017).
13 hours wer
were spent on an NVIDIA
IA Tesla P100
00 GPU during
ring training. An illustration
n of reconstru
truction outcom
omes
comparing co
convolutionall networks
n
train
rained with clea
clean and noisyy ttargets, respect
pectively, is pres
resented in Figu
igure 9(b, c). Our
O
results are
re pre
pretty similarr to those publis
blished in priorr sstudies in term
erms of PSNR.
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 47
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
MERITS
There is no
o doubt about
ut it; the followi
llowing are the
he advantages
es of
o the sugges
gested approach
ach for picture
re reconstructi
ction:
Effective met
ethods for reducing
red
noise:
e: T
Through thee uutilization of the suggested
ted technology,
gy, noise from d
degraded pho
hotos
may be succes
ccessfully reduced
uced, resulting
ng in improved
ved visual clarity
ty and enhanced
ced detail. Thee significance
ce o
of this cannot
ot be
overstated w
when it comes
es to application
tions in medical
cal imaging, where
here noise can
n conceal crucial
cruci diagnostic
ic information.. The
Th
removal off ar
artifacts: Images
ages that have
ve bbeen deteriora
iorated can have
ave artifacts such
suc as blurring
rring, aliasing, and moiré pattern
tterns
successfullyy re
removed usin
sing the metho
ethods that hass been provided
vided. The preserva
reservation off the
th accuratee portrayal
p
off the
underlying im
image and the preventionn of misunders
erstandings are both major
jor reasons why
wh this is crucial.
cru
This is an
improvement
ent in resolutio
tion: The appro
proach that ha
has been devel
eveloped has the
th potentiall to
t improvee the
th sharpness
ess of
photographs
hs and disclose
se finer details
ails, thereby increa
increasing thee resolution
res
of reconstructed
cted images. This
T
is useful
ul for
applicationss in astronom
my, which is a field in wh
which the cap
capacity to recognize
reco
minu
inute features
res is essential
al for
comprehendi
nding the cosmo
smos. Low complexity
com
in ter
terms of compu
putation: Thee suggested technique
tec
hass a low comput
puting
complexity,, w
which makes
kes it possiblee to
t do proces
cessing in real time or very close to real
rea time situat
uations. Thesee are
applicationss tthat demand
d quick picture
ure reconstruct
uction, such as medical imag
aging and survei
rveillance, and
d thus is a cru
rucial
componentt ffor such appl
plications. The memory foot
ootprint has been
bee reduced fo
for: Processing
ing of big pictu
cture collections
ons is
made possib
sible by the suggested
su
meth
ethodology, wh
which has a smaller
sm
memo
ory footprint
nt than other
er aapproaches.
es. For
F
applicationss tthat deal with
wit enormous
us amounts of picture data
ata, such as tho
those found in digital archive
chives and satel
tellite
photography,
hy, this is a signi
gnificant considera
ideration.
Algorithmss th
that can bee parallelized
pa
In
n order
o
to facili
cilitate efficient
nt implementati
tation on multi--core processo
essors and graph
raphics
processing un
units (GPUs),
), the
t suggested
ed technique ma
makes use of pa
parallelizablee algorithms.
alg
Con
onsequently,
y, th
this speeds up
p the
process off ppicture reconstruction
recon
eve
even further
er aand makes it easier to
o analyze enor
ormous dataset
asets. Capacity
ity to
accommodate
ate a wide variety
variet of data types:
typ The app
pproach that has
ha been develo
eveloped is flexib
exible enough to accommodat
date a
wide variety
ety of data form
rmats and perf
erforms well acro
across a widee range of pict
icture kinds, including
in
astro
tronomical phot
hotos,
medical imag
ages, and nature
ture photograph
raphs. As a resul
result of its adapt
aptability, it is aan extremely
ely useful instrum
rument for a wide
wi
variety of ap
applications. Imaging
Im
techni
hniques that are resistantt tto its effects:
ects: The suggest
ested technique
que is resilient
nt to
fluctuationss in imaging modalities,
mo
inclu
cluding X-rays,
ys, m
magnetic reso
resonance imagin
ging (MRI), computed
com
tomog
ography (CT),
T), aand
astronomical
cal observatories
ries. Becausee o
of this, it iss a trustworthy
rthy option forr image reconstruction
recon
acro
cross a variety
ety of
imaging hardw
rdware and soft
oftware. Imagee features
f
that
at are sensitive
ve to the followi
wing: The appro
proach that has
as been develop
loped
is sensitive
ve to the particula
cular aspects off the
t picture, su
such as the tex
texture, contras
rast, and noisee patterns
p
thatt aare present.. As
A a
result, it is aable to modi
odify the recon
reconstruction pro
process so that
hat it is tailored
red to the part
articular image
ge content, wh
which
ultimately lea
leads to superio
erior outcomes.
es. Preservation
on of photograp
raphic characteri
cteristics While
le the reconstru
truction proces
cess is
being carried
rried out, the suggested
sug
techn
echnique ensures
res that essenti
ential picture ch
characteristics,
cs, such as edges
ges, textures,
res, aand
anatomicall st
structures, are
re maintained.
ed. T
The integrity
ty of the recons
reconstructed pictu
cture must bee preserved
p
att all
a costs, and
d tthis
is absolutely
ly necessary in order to guarantee
gua
appro
ropriate interpre
erpretation. Accu
ccuracy in matt
atters pertaining
ing to quantitat
itative
measurement
ents: In order
er to get precise
se quantitative
ve m
measures from
rom rebuilt pictu
ctures, the appro
proach that has been presen
ented
is utilized. Bec
Because of this,
his, it is possible
ble to conduct
ct a trustworthy
rthy analysis and
nd interpretatio
tion of quantitat
itative picture
re da
data,
which is a ke
key componen
nent for applica
plications in sci
scientific research
resea
and med
edical imaging.
ng. The interp
erpretability off the
methodology
gy of reconstr
struction: Thee techniquee th
that has been
een suggested offers
o
manyy insights
i
into
to the process
cess of
reconstructio
ction, which enables
ena
a better
etter knowledge
ge of the rebui
rebuilt picture as well as thee constraints that
th it possess
esses.
Because of th
this interpretab
tability, trust in the reconstru
struction proces
cess is increased
sed, and it is easier
ea
to make
ke decisions based
bas
on accurate
te information.
n. In
I a nutshell,
ll, the suggested
ted approach for
fo picture reco
reconstruction
n provides a compelling
co
mix
ix of
benefits, whic
which makes itt an
a invaluablee instrument
in
for a broad variety
vari
of applica
lications. Thee fact
fa that it iss aable to efficien
iently
decrease noi
noise, eliminate
te artifacts, imp
improve resolut
lution, and reta
retain picture ch
characteristics
cs while keeping
ing a low level
vel of
computationa
onal complexity
ity and flexibilit
ility makes itt a viable techn
hnique for ima
mage restoratio
tion and improve
rovement. pictu
cture,
including edg
edges, textures,
es, and anatomica
ical structures,
es, while reconstructing
recon
the
he image. Reliable
Reliab analysis an
and interpretat
tation
are made po
possible by the
he accurate qua
uantitative meaasures that are obtained fro
from the rebui
ebuilt pictures.. In
Interpretability
lity of
Reconstructio
ction Process: Gives
G
informat
ation on how
w tthe picture
re was
wa rebuilt and its limits, enabling
en
a better
etter understand
nding
of the proces
cess.
DEMERITS
TS
Although the suggested approach
a
for
or picture reco
reconstruction ha
has several en
encouraging benefits,
ben
there
re are also cert
certain
disadvantages
ges that should
ld be taken into
nto account: De
Depending on
n how
h
it is impl
plemented speci
pecifically and the
th kind of image
im
that needs to be reconst
nstructed, thee ssuggested me
methodology for
fo image reco
reconstruction may
m have som
ome drawbacks
ks or
restrictions. Several possib
ssible drawbacks
cks consist of:f: Amplification
on of Noise Important
Imp
inform
ormation mayy be
b obscured
red and
a
picture qualit
ality may be decreased
decre
if the
he restorationn pprocedure ad
adds additional
al noise or amp
mplifies already--present noise
ise in
the damaged
ed image. Introd
roduction to th
the Artefact D
During the reconstruction
reco
n process,
p
new
ew artefacts like
ike aliasing, ring
inging,
and blurring
ng might appear.
ear. These wou
would lower the
he overall qualit
ality of the pict
icture and migh
ight have an impact
im
on furth
rther
analysis or in
interpretation.
n. Resolution Restrictions
Res
Th
The intrinsicc res
resolution off the
t degraded
ed data may restrict
res
the feasi
asible
resolution of the reconstr
structed image,
ge, making it im
impossible to
o reco
recover tinyy features and
d thus jeopardi
rdizing the imag
age's
suitability for some uses.. Computation
ional Overhead
ead When worki
rking with huge
ge picture data
atasets, complex
lex reconstruct
ruction
techniques m
may need a lot of computati
ational resources
rces, such proces
cessing power
wer and
a memory,
ry, which makes
kes them unsuita
itable
for real-timee or nearly real-time
rea
applica
ications.
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 48
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
Dependency
cy on Data The
Th qualityy an
and features
res of the deter
eteriorated data
ata may have
ve a significant
nt impact on
n the
reconstructio
ction method's
's performance.
ce. Diverse da
datasets mayy ex
exhibit incons
consistent recon
econstruction qua
quality because
se to
differences inn noise levels,
evels, picture content,
cont
and data
ata gathering settings.
se
Reliabi
iability to Extrem
reme Parameters
eters The choice
ice of
hyperparamet
meters, or varia
riables that regulate
regu
the beh
ehavior of thee reco
reconstructio
ction algorithm,, may
m have an impact on there
th
construction
on process. To get the best
est results, caref
careful tweaking
king and experie
erience are freq
requently needed
eded, and the id
ideal
collection of hyperparamet
meters might change
ch
based
ed o
on the particu
rticular picture
re and
a application
tion. Computati
ational complex
lexity:
Reconstructio
ction techniques
ues based on deep learning m
may be compu
putationally cos
costly, particularl
larly when dealin
ealing with comp
mplex
degradationn sscenarios orr high-resolutio
h
tion images. Th
This may restrict
trict their usee in real-timee applications
ap
or on devicess with
wi
limited resou
ources. Requirem
irements for data:
dat
Training deep
eep learning mo
odels for image
age reconstruct
ction frequently
ently calls for a substantial
su
volume
volu
of excelle
ellent training da
data,
which can be challengingg to come byy for particular
lar kinds of images
ima
or degra
egradation scena
enarios. Over fit
fitting and sub
ubpar
performance
ce on unobserved
erved data mayy res
result from th
this. Noise sen
ensitivity: Artefa
rtefacts or erron
rroneous reconstr
structions in noisy
no
images mayy rresult from deep
dee learningg models’ sensit
nsitivity to noise
ise in the traini
ining set. Efficien
ciency of Comp
mputation: Minim
inimal
computationa
onal resources
rces are needed
ed for real-time
me or almostt rea
real-time pro
rocessing, which is especial
ecially beneficial
al for
dynamic imag
aging applicatio
tions. This is kn
known as low
w computationa
nal complexity.
ity. Diminished Memory Foot
otprint: Effective
ectively
employs mem
emory, enabling
ling the process
cessing of substa
stantial picture
re collections and
an diminishing
ing the necessit
essity for hardwa
ware.
Algorithms th
that can bee parallelized
pa
are
re used, which
ch allows forr the
th effective us
use of multi-co
core CPUs and
nd GPUs for rapid
ra
reconstructio
ction. Applicabili
bility in general
eral: Flexibilityy tto Varying Data
Da Types: Wo
Works effective
ectively with a variety
va
of pictu
icture
formats, such
uch as astrono
nomical, medica
ical, and Interp
erpretability Difficulties
Di
It can
ca be difficu
icult to compre
prehend the inner
inn
workings of complicated
ted reconstruct
uction algorithm
rithms, especiall
ecially those that
tha rely on deep
d
learning
ing. This lack
ck of
interpretabilit
bility may imped
ede confidence
ce in the recon
reconstruction pro
rocess by makin
aking it challengi
enging to evaluat
uate the limitatio
ations
and dependab
dability of thee reconstructed
reco
ed image.
C
CONCLUSI
SION
Throughoutt this extensive
sive literature analysis,
a
a vari
variety of image
age restoration
n techniques have
h
been investigated.
inves
Th
These
techniques ra
range from the
th more conve
nventional filter
lter-based appro
proaches to the most cutting
ting-edge deep
ep learning mod
odels.
While there
ere are certain circumstances
ci
ces in which trad
raditional appro
roaches have
ve bbeen shown to be efficient,
ent, they frequen
uently
lack flexibility
ility and suffer
er when
w
it comes
es to accomplis
lishing complica
licated degrading
ing jobs. Deep learning, and
d more specifica
cifically
convolutional
nal neural netwo
etworks (CNNs
Ns), has emerg
erged as a pot
potent techniqu
ique for image
ge restoration.
n. It has achieve
ieved
outstandingg ssuccess in eliminating
elim
noise
ise, blurring, dis
distortions, and
nd other impairm
airments from images.
im
Genera
nerative adversa
versarial
networks, oft
often known as GANs, have
ve demonstrated
rated further pro
promise in succes
ccessfully addres
ressing complica
licated restorat
ration
issues, particu
rticularly in situat
uations when data
da is lacking.
g. Image restora
oration is stilll facing
fa
a number
ber of seriouss obstacles,
o
desp
espite
the substantia
ntial progress that
th has beenn m
made in thee fiel
field: Deep lear
earning models
els frequently function
fun
as "black
lack boxes," which
wh
makes it chall
hallenging to comprehend
com
thei
their inner worki
rkings and deci
ecision-makingg procedures.
p
This is one
eo
of the reaso
sons why thei
eir interpreta
etability is limi
mited.
1. Depende
dence on dat
ata: In order
er tto train succes
ccessful deep
eep learning
lea
model
dels, it is neces
ecessary to gath
ather a substan
tantial
quantityy o
of high-quality
lity data, whichh m
may be bothh ex
expensive and
nd time-consum
uming to acquire.
uire.
2. The expe
pense of comp
mputation: The
Th process of training and
nd deploying deep
dee learning models
m
can be computation
onally
costly, neces
necessitating the use of speci
ecialized hardwa
ware and softwa
tware resources
rces.
3. Generaliz
alizability: Mod
odels that have
ave been trained
ined on certain
in datasets coul
could not genera
eralize well to d
data that has
as nnot
been seen
een before orr to
t different fo
forms of deteri
eterioration. In th
the future, futu
future research
rch in picture res
restoration sho
hould
concentra
rate on finding
ing solutions to these proble
blems by develo
veloping: Increasi
asing the number
ber of interpret
pretable models
els by
employing
ying methods such
su as explain
lainable artificial
cial intelligence
ce (XAI) to com
comprehend the behaviorr of models and
nd to
enhance
ce co
confidence. Methods
M
that
at are effective
ve with data inc
include employin
ying strategies
es such as trans
ransfer learningg aand
data augm
ugmentation in order to lessen
les
the de
dependency on lengthy dat
datasets. Invest
estigating effect
ective designss aand
optimizati
ation strategies
ies in order to reduce thee aamount of com
computationall resources
res
req
equired for lightweight
ligh
mod
odels.
The devel
evelopment off models
m
that are generalizabl
able over a wide range of degradation
deg
kind
inds and domai
ains is referred
erred to
as domain
ain-agnostic modeling.
mo
Picture
cture restoration
ion technology
gy can continue
ue to advance
ce if these issues
es are address
ressed.
This will
ill allow for a wider
wi
rangee of
o applications
ns in a variety
ety of fields, such
uch as medical
cal imaging, rem
emote sensing,
g, and
a
autonomo
mous vehicles,
es, which willl ultimately
u
lead
ead to an impro
provement in
n picture
p
comp
prehension an
and interpretat
tation
across a va
variety of field
elds.
REFERENC
NCES
[1]. Mei, Y., Fan, Y., Zha
hang, Y., Yu,, JJ., Zhou, Y., LLiu, D., ... & Shi, H. (202
023). Pyramid
d attention
a
netw
etwork for ima
image
restorat
ration. Internati
ational Journal
al o
of Computer
er Vi
Vision, 131(12
(12), 3207-3225
25.
[2]. Lehtinen
nen, J., Munkberg
kberg, J., Hasselg
elgren, J., Laine,
ne, S., Karras,, T.,
T. Aittala, M., & Aila, T. (2018).
(2
Noise2
e2Noise: Learni
rning
imagee res
restoration without
wit
clean da
data. arXiv pre
reprint arXiv:18
:1803.04189.
[3]. Zamir,
r, SS. W., Arora,
ra, A., Khan, S., Hayat, M., Kh
Khan, F. S., Yan
Yang, M. H., & Shao,
S
L. (2020
20). Learning en
enriched features
tures
for rea
real image restoration
rest
and
d enhancemen
ent. In Comput
puter Vision–EC
ECCV 2020:: 16th
1
Europea
pean Conferen
erence,
Glasgow,
ow, UK, August
ust 23–28, 2020
20, Proceedings
gs, Part XXV
V 16 (pp. 492-511
511). Springer
er International
In
l Publishing.
P
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 49
IJIRIS: Inter
ternational Jour
urnal of Innovat
vative Research
rch in Informatio
tion Security
Volume 10,
0, Issue 02, February
Feb
2024
E-IS
ISSN: 2349-701
7017
P-IS
ISSN: 2349-700
7009
https://www.ijiri
ht
.ijiris.com/archiv
hives
[4]. Zhang,
g, Y., Tian, Y.,., Kong, Y., Zh
Zhong, B., & FFu, Y. (2020)
0). Residual den
dense network
rk for imagee restoration.
res
IE
IEEE
transact
actions on patter
ttern analysis an
and machinee int
intelligence, 43((7), 2480-2495
495.
[5]. Zhang,
g, K
K., Li, Y., Zuo
uo, W., Zhang,
g, L., Van Gool
ol, L., & Timoft
ofte, R. (2021).
). Plug-and-play
lay image restora
toration with deep
de
denoiser
iser prior. IEEE
E Transactions
Tra
ns on
o Pattern An
Analysis and Ma
Machine Intelligen
lligence, 44(10),
), 6360-6376.
[6]. Zhang,
g, Y., Li, K.,., Li, K., Zhon
Zhong, B., & Fu
Fu, Y. (2019).
9). Residual non-local
n
atte
ttention netwo
works for ima
image
restorat
ration. arXiv preprint
pre
arXiv:1
iv:1903.10082.
[7]. Dong,, W.
W., Wang, P.,
P. Yin, W., Sh
Shi, G., Wu, F., & Lu, X. (2
(2018). Denoisi
oising prior driven
riven deep neu
eural network
rk ffor
imagee res
restoration. IEEE
IEE transaction
ions on pattern
rn analysis and
d machine
m
intell
elligence, 41(10
(10), 2305-2318.
8.
[8]. Delbrac
racio, M., &Mila
ilanfar, P. (2023
023). Inversionn by direct itera
iteration: An alternative
alt
to
o denoising
d
diffu
iffusion for ima
image
restorat
ration. arXiv preprint
pre
arXiv:2
iv:2303.11435.
[9]. Luo, Z., Gustafsson,
n, F. K., Zhao,
ao, Z., Sjölund,
d, J., & Schön,
n, T. B. (2023).
23). Image restoration
rest
with mean-reverti
everting
stochast
astic differential
tial equations. arXiv
arX preprint
nt arXiv:2301.11
.11699.
[10]. Glaubitz
itz, J., Gelb,, A.,
A & Song,
g, G. (2023).
). Generalized
ed sparse Bayes
Bayesian learning
ing and applica
lication to image
ima
reconstr
struction. SIAM
M/ASA Journal
rnal on Uncertain
ainty Quantifica
ication, 11(1),, 26
262-284.
__________
_____________
______________
______________
_____________
_____________
______________
______________
_____________
____
IJIRIS © 2014-24, AM
M Publications
ns -All Rights Res
Reserved
https://doi.org
org/10.26562/ijiri
/ijiris
Page- 50