In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
DOI :10.5121/ijai .2016.6103 35
A COMPARATIVE STUDY ALGORITHM FOR NOISY
IMAGE RESTORATION IN THE FIELD OF MEDICAL
IMAGING
D .P.Sumi a
Assis an P o esso , Depa men o Compu e Science, Vi ekanandha College o A s and
Sciences o Women(Au onomous)
Elayampalayam,Ti uchengode
A
BSTRACT
This pape p esen s he pe o mance analysis o di e en basic echniques used o he image es o a ion.
Res o a ion is a p ocess by emo ing blu and noise om image and ge back he o iginal o m. Medical
images play a i al ole in dealing wi h he de ec ion o a ious diseases in pa ien s and hey ace he
p oblem o sal and peppe noise and Gausian noise. Hence es o a ion is pe o med based on di e en
image es o a ion echniques. In his pape , popula es o a ion echniques is applied and analyzed in he
eco e y o medical images,.
K
EYWORDS
Image Res o a ion, In e se Fil e , Weine Fil e , MSE, PSNR, Medical Images
1.
I
NTRODUCTION
Image P ocessing is a o m o signal p ocessing o which he inpu is an image ei he i can be
pho og aph o ideo ame. The ou pu o he image can be se o cha ac e is ics o pa ame e s
ela ed o he image. I includes se e al echniques such as image segmen a ion, image
ecogni ion, image es o a ion, e c[1]. Image Res o a ion is an up-and-coming ield o image
p ocessing which e e s o g oup o me hods o echniques ha ocus on eco e ing an o iginal
image om a deg aded image. The deg ada ion may occu due o se e al ways ha includes
senso noise, came a mis ocus, ela i e objec -came a mo ion, andom a mosphe ic u bulence
[2][3]. The e a e wo subp ocesses.
i. Deg ading he quali y o he image by adding blu and noise o an image
ii. Reco e ing he o iginal image.
Deg ada ion Model o Blu ing Image
In deg ada ion model o blu ing image, he image is blu ed using di e en kinds o il e s and
an addi i e noise. The image can be deg aded by using sal and peppe noise and Gaussian Noise.
The deg aded image can be exp ess by he equa ion as
= g * h + n; (1)
whe e deno es he obse ed blu ed image
g deno es he clea image o eco e
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
36
h deno es he blu ke nel
n deno es he noise
The below Figu e 1 ep esen s he o ma ion o deg ada ion model o blu ing image.
Figu e 1. Deg ada ion Model o blu ing image
Res o a ion Model
Res o a ion is ob ained by deg ading he image using es o a ion il e s. In his p ocess, noise and
blu image ac o is emo ed and we ge he es ima ed o iginal image. Figu e 2 ep esen s he
s uc u e o es o a ion model [4].
Res o a ion Fil e
Inpu Image
Ou pu Image
Figu e 2: Res o a ion Model
2.
R
ESTORATION
T
ECHNIQUES
The wo ypes o image es o a ion echniques a e
1. Blind Techniques
2. Non-Blind Techniques
Inpu
Image
O iginal
Image +
Deg aded
Func ion
Noise
Deg aded
Image
Ou pu
Image
Deg aded
Image ( In e se Fil e
/ Weine
Fil e )
Res o ed Image
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
37
2.1 Blind Image Res o a ion
This me hod allows he econs uc ion o o iginal images om deg aded image. The majo ype
applied is Blind Image Decon olu ion(BID). I is di icul o implemen and mo e complica ed
when compa ed o o he ca ego y[5]
2.2 Non-Blind Res o a ion
This echnique helps in econs uc ion o o iginal image om deg aded image. The image can be
deg aded by ha ing he knowledge o PSF. In his pape we a e going o conside abou In e se
Fil e ing and Weine Fil e ing.
3.
INVERSE
F
ILTERING
In e se il e ing is he quickes and easies way o es o e he blu ed image. Blu ing can be
conside ed as low pass il e ing in in e se il e ing app oach we use high pass il e ing ac ion o
econs uc he blu ed image wi hou much e o .
4.
W
EINER
F
ILTERING
Weine il e ing has been a classical ool in signal p ocessing and communica ion since he
1950’s[6].I is a s anda d image es o a ion app oach o bo h he deg aded unc ion and
s a is ical cha ac e is ic o noise in o he es o a ion unc ion. I is non-blind echnique o
econs uc ing he deg aded image which is known as PSF. I emo es he addi i e noise and
in e s he blu ing simul aneously. Wiene il e no only pe o ms he decon olu ion by in e se
il e ing bu also emo es he noise wi h he comp ession ope a ion. The inpu o a weine il e
is a deg aded image co up ed by addi i e noise [7]
5.
M
EDICAL
I
MAGE
R
ESTORATION
Image es o a ion echniques a e applied o es o e images om as onomical ield and sa elli e
images [8][9]. In his pape , medical image es o a ion is mainly ocused.
5.1 Medical Images
Medical Images a e used o de ec a numbe o diseases which canno be de ec ed anywhe e.
These images may be con amina ed wi h noise o blu image which eels di icul o he doc o
o iden i y he diseases. This es o a ion o such deg aded images is mus o he well being o
he common people. The e a e a numbe o medical images which a e X- ay images, Ul asound
images, CT scan images and MRI images as shown Figu e 3.
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
38
Figu e 3: Medical Images
5.1.1 X- ay images
I is a popula imaging es ha has been used o decades o help doc o s iew he inside o he
body wi hou ha ing o make an incision. I is used o ind o hopaedic damage, umou s,
pneumonias, e c.
5.1.2 Ul asound images
Ul asound image uses high- equency sound wa es o iew inside he body. Ul asound a e
cap u ed in eal- ime, hey can also show mo emen o he body’s in e nal o gans as well as blood
lowing h ough he blood essels. I is a medical ool ha can help a physician e alua e,
diagnose and ea medical condi ions.
5.1.3 CT Scan images
Compu ed omog aphy(CT) is a nonin asi e medical examina ion ha uses specialized X- ay
equipmen o p oduce c oss-sec ional images o he body. CT scans can be pe o med on e e y
egion o he body o a a ie y o easons (e.g., diagnos ic, ea men planning, in e en ional, o
sc eening). Mos CT scans a e pe o med as ou pa ien p ocedu es.
5.1.4 MRI images
Magne ic esonance imaging(MRI) is a medical imaging p ocedu e ha uses s ong magne ic
ields and adio wa es o p oduce c oss-sec ional images o o gans and in e nal s uc u es in he
body. Using MRI scans, physicians can diagnose o moni o ea men s o a a ie y o medical
condi ions which includes umo s, hea p oblems, diseases o he li e e c.
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
39
6.
R
ESULTS
F
OR
D
EBLURRED
I
MAGES
The p oposed app oach is expe imen ed using a es image X- ay, CT, MRI,and Ul asound
image o size 256x256 as shown in Figu e 5,Figu e 6, Figu e 7 and Figu e 8 wi h di e en noise
ype
.
Figu e 5. O iginal Image o Medical Imaging o Size 256 x 256
Figu e 6. Deg aded Image o Medical Imaging o Size 256 x 256 wi h di e en Noise ypes
Figu e 7. Res o ed image o X-Ray Ches , CT, MRI and Ul asound using In e se Fil e
Figu e 8. Res o ed image o X-Ray Ches , CT, MRI and Ul asound using Wiene Fil e
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
40
7.
P
ERFORMANCE
P
ARAMETERS
Fo pe o mance pa ame e s, a ious medical images we e conside ed o di e en ype and size
256 X 256. Pe o mance Me ics a e MSE (Mean Squa e E o ) and PSNR (Peak Signal o Noise
Ra io).
MSE: I is de ined as he di e ence be ween he ac ual and he es ima ed signals . I is exp essed
as ollows.
MSE = ]
2
(2)
Whe e I,K ep esen he o iginal image and he es o ed images espec i ely o size mxn. An i,j
ep esen he pixels o images.
PSNR: I is de ined as he a io be ween he maximum possible powe o a signal and he powe
o compu ing noise ha a ec s he ideli y o i s ep esen a ion. PSNR is exp essed in e ms o
decibel (dB). I is exp essed ma hema ically as ollows:
PSNR = 10.log
10
( ) (3)
8.
E
XPERIMENTAL
R
ESULTS
In his pape we ha e he i e a i e esul s in he o m o deblu ed image, PSNR and MSE alues
o Blind decon olu ion algo i hm. When we se he i e a ions o 50 in Table 1 o ou
expe imen , we ha e he ou pu image along wi h he PSNR alue a maximum o 29.61852dB
and MSE+0.00109 dB o Weine il e . Then as we inc ease he numbe o i e a ions o 70 as
shown in Table 2, he PSNR alue g adually dec eases when compa ed o p e ious i e a ions a a
maximum o 29.43213dB and MSE +0.00114 dB o Weine il e . I can be ep esen ed
g aphically in he o m o g aph so ha we can isualize i in an easy manne . The Figu e 9.
below shows he compa ison o in e se il e and weine il e o an i e a ion 50.
Table 1. MSE & PSNR Values o Medical Images o size 256 x 256 a an I e a ion o 50
S.N
o Image
Pe o mance Measu es
In e se Fil e Weine Fil e
Mean Squa e
E o
(MSE)
Peak Signal o
Noise Ra io
Mean Squa e
E o
(MSE)
Peak Signal o
Noise Ra io
1 X-Ray
Ches .png +0.01716 dB
+17.65530 dB
+0.00165 dB +27.82729 dB
2 CT.jpg +0.00810 dB +20.91485 dB +0.00397 dB +24.00683 dB
3 MRI.jpg +0.02057 dB +16.86871 dB +0.00109 dB +29.61852 dB
4 Ul asound.
png +0.00750 dB +21.24827 dB +0.00403 dB +23.94484 dB
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
41
Table 2. MSE & PSNR Values o Medical Images o size 256 x 256 a an I e a ion o 70
S.No Image
Pe o mance Measu es
In e se Fil e Weine Fil e
Mean Squa e
E o
(MSE)
Peak Signal o
Noise Ra io
Mean Squa e
E o
(MSE)
Peak Signal
o Noise
Ra io
1 X-Ray
Ches .png +0.01744 dB +17.58538 dB +0.00170 dB +27.69042
dB
2 CT.jpg 0.00810 dB +20.91277 dB +0.00406 dB +23.91201
dB
3 MRI.jpg +0.02089 dB +16.80152 dB +0.00114 dB +29.43213
dB
4 Ul asound.p
ng +0.00740 dB +21.30796 dB +0.00393 dB +24.05538
dB
(a) (b)
Figu e 9. Compa ison o In e se il e and Weine il e o an i e a ion 50.
9.
C
ONCLUSIONS AND
F
UTURE
W
ORK
In his pape , we ha e s udied a me hod o blind decon olu ion algo i hm. To conclude
es o a ion o medical images like X- ay, CT, MRI and ul a-sound is e y c ucial and sensi i e
ma e . The es o a ion mus be done o maximum possible le el. The PSNR alue o In e se
il e is poo when compa ed wi h ha o Weine Fil e . Fu u e wo k o his pape is o de elop
and build imp o ed echnique which will gi e be e pe o mance han Weine .
R
EFERENCES
[1] Cha u Kha e and Kapil Kuma Nagwanshi,”Image Res o a ion Technique wi h Non Linea Fil e ”,
In e na ional Jou nal o Ad anced Science and Technology, Vol.39,pp. 67-74,Feb ua y 2012.
[2] Rinku Kalo a , Sh.Anil Saga ,”A Re iew: A No el Algo i hm o Blu ed Image Res o a ion in he
ield o Medical Imaging”,In e na ional Jou nal o Ad anced Resea ch in Compu e and
Communica ion Enginee ing,Vol.3, Issue 6, June 2014.
In e na ional Jou nal o Ad anced In o ma ion Technology (IJAIT) Vol. 6, No. 1, Feb ua y 2016
42
[3] Anamika Mau ya, Rajinde Tiwa i,”A No el Me hod o Image Res o a ion by using Di e en Types
o Fil e ing Techniques”, In e na ional Jou nal o Enginee ing Science and Inno a i e
Technology(IJESIT) , Vol 3, Issue 4, July 2014.
[4] Amandeep Kau Vinay Chop a, “Blind Image Decon olu ion Technique o Image Res o a ion Using
An Colony Op imiza ion,”In e na ional Jou nal o Compu e Applica ions & In o ma ion Technology
Vol. 1, Issue 2, Sep embe 2012.
[5] Ramya, S.; Me cy Ch is ial, T, “Res o a ion o blu ed images using Blind Decon olu ion
Algo i hm,” IEEE, Eme ging T ends in Elec ical and Compu e Technology (ICETECT),
pp.496 –499,2011.
[6] Ande s Ahlen and Mikael S e nad,”Wiene Fil e Design Using Polynomial Equa ions”, IEEE
T ansac ions on Signal P ocessing,Vol.39,No.11, pp.2387-2399,No embe 1991.
[7] h p://dip.sun.ac.za/~m ma i z/DIP/DIPweb/ Wiki_Wiene _decon olu ion.pd .
[8] Sanchez MG, Vidal V, Ve du G, Mayo P,Rodenas F.,”Medical Image Res o a ion wi h di e en ypes
o noise”,IEEE Eng Med Biol Soc. Con e ence P oceedings 2012.
[9] P.Sugan hi, M s.A.Gokila, D .K.Ramasamy,”Sa elli e Image Res o a ion Using Shea le
T ans o m”,In e na ional Jou nal o Scien i ic & Enginee ing Resea ch, Vol.5,Issue 5, May-2014
Au ho
D .P.Sumi a. ecei ed he Ph.D Deg ee in Compu e Science om Mo he Te esa
Women’s Uni e si y, Kodaikannal, Tamil Nadu, India in he yea 2013. She is
cu en ly wo king as a Assis an P o esso in Depa men o Compu e Science,
Vi ekanandha College o A s and Sciences o Women, Elayamapalayam,
Ti uchengode, TamilNadu,India. She published 17 In e na ional Jou nal pape s, 4
pape s in In e na ional Con e ence and 10 pape s in Na ional Con e ences. He
esea ch a eas include Image P ocessing, BioMe ics, Da a Mining and A i icial
In elligence. She has 13 yea s 8 Mon hs o eaching expe ience in sel inance
ins i u ions.
