DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
BEMD Based Ul asound Image Speckle Reduc ion
Technique Using Pixel-Wise Wiene Fil e ing
Bhawna GUPTA , Vinee KHANDELWAL
Depa men o Elec onics and Communica ion Enginee ing, Jaypee Ins i u e o In o ma ion Technology,
Sec o -62, 201309 Noida, U a P adesh, India
bha[email p o ec ed], inee .khandelw[email p o ec ed]
DOI: 10.15598/aeee.19i2.4100
A icle his o y: Recei ed Jan 30, 2021; Re ised Ma 17, 2021; Accep ed Ap 01, 2021; Published Jun 30, 2021.
This is an open access a icle unde he BY-CC license.
Abs ac . In his pape , an imp o ed Bidimensional
Empi ical Mode Decomposi ion (BEMD) based speckle
educ ion echnique o ul asound images has been p o-
posed. The noisy image has been decomposed in o i s
In insic Mode Func ions (IMFs) and a esidue. The
noise componen o he low o de IMFs is emo ed wi h
he pixel-wise Wiene il e ing. The image is econ-
s uc ed wi h hese il e ed low o de IMFs, high o de
IMFs and he esidue. The pe o mance o he p o-
posed me hod has been es ed on syn he ic as well as
eal ul asound images ha ing noise componen s o di -
e en a iance. The expe imen al esul s show ha he
p oposed algo i hm pe o ms be e han o he exis ing
me hods o syn he ic images as well as eal ul asound
images in e ms o a ious image quali y ma ices.
Keywo ds
BEMD, noise educ ion, speckle, ul asound,
Wiene .
1. In oduc ion
Speckle, which is a mul iplica i e noise, is an unwan ed
phenomenon ha is p esen in ul asound images
due o sca e ing a he ime o acqui ing he image
[1]. Speckle educ ion in ul asound images is an essen-
ial s ep and is a ge ing imp o emen in he quali y
o he image in e ms o PSNR (Peak Signal o Noise
Ra io), CC (Co ela ion Coe icien ), SNR (Signal
o Noise Ra io), FoM (Figu e o Me i ) and SSIM
(S uc u al SIMila i y) [2]. Due o he high- equency
cha ac e is ic o he noise componen , he main chal-
lenge o speckle educ ion echnique is ha while e-
mo ing noise he in o ma ion in he edges should no
be los as his is impo an o diagnosis. The me ic
used o measu ing he same is he EKI (Edge Keeping
Index).
Image denoising speci ically speckle educ ion in ul-
asound images has been s udied ex ensi ely, and
i has been b oadly classi ied in o di e en domains,
such as spa ial domain echniques, ans o m do-
main echniques, and hyb id echniques as shown
in Fig. 1. The spa ial domain echniques wo k di-
ec ly on he image pixels while ans o m domain
echnique applies an app op ia e ans o m o con-
e he image o equency domain be o e p o-
cessing. Fu he mo e, he e a e hyb id echniques
ha a e combina ion o spa ial domain and ans o m
domain me hods.
The spa ial domain echniques use local s a is ics
o in o ma ion edundancy be ween simila pa ches
and eplace he pixel alue by p ocessing he nea by
pixel alues. Mos success ul amongs his ca e-
go y a e di usion-based il e s like Speckle Reduc-
ing Aniso opic Di usion (SRAD), De ail P ese -
ing Aniso opic Di usion (DPAD), Pe ona-Malik’s
Aniso opic Di usion (PMAD) [3], [4], [5] and [6], Bi-
la e al il e s [7] and [8], and pa ch-based me hods like
Non Local Mean il e (NLM) [9], [10], [11] and [12] and
Op imized Bayesian Nonlocal Mean il e (OBNLM)
[13] and [14]. The simila pa ch-based me hods need
o selec he candida e pa ch easonably so ha he e-
mo al o noise does no lead o lose o edge in o ma ion.
The e o e, ecen wo k as p oposed in [15], [16] and [17]
uses modi ied NLM and BM3D algo i hms while y-
ing o p ese e he edge in o ma ion. Wiene il e s
[18] and [19] a e he op imum linea il e s, ha a e
widely in use o image p ocessing. Pe o mance o he
classical poin wise Wiene il e is enhanced o noise
168 ©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
Fig. 1: Classi ica ion o Speckle educ ion echniques.
co up ed images, wi h he use o non-local pa ame e
es ima ion p oposed by And e e al. [20].
T ans o m domain me hods assume ha he im-
age can be spa sely ep esen ed by i s low- equency
componen s and he speckle is p esen in he high-
equency componen s o he image [21]. The eby, he
image high- equency componen s a e deal wi h by ap-
plying he h esholding (ha d o so ) as in wa ele
h esholding me hod o emo e he noise [22]. While
hese me hods gi e sa is ac o y esul s, choosing he
igh h eshold is icky and should a oid he Gibbs
phenomenon and o e smoo hening e ec s. Thus, u -
he ex ension in he ans o m domain echniques
includes da a-d i en echniques and modeling ech-
niques.
Hyb id echniques a e he ones ha combine spa ial
domain me hods wi h ans o m domain me hods o
he imp o emen o denoising pe o mance. Some o
he ecen hyb id echniques a e as in oduced in [23],
[24] and [25].
Empi ical Mode Decomposi ion (EMD) was in o-
duced by Huang e . al [26] in 1998. This is a e y
use ul algo i hm ha decomposes he signal in i s In-
insic Mode Func ions (IMFs) [27]. This decomposi-
ion is done based on hei local equency o oscilla-
ion in he spa ial domain. Unlike Fou ie ans o m
and Wa ele ans o ms, he basis unc ions calcula ed
a e signal-dependen and a e used o es ima e a se ies
o IMFs ia an i e a i e p ocedu e called si ing [28].
EMD was in oduced in images in 2003 [29] and
he Bidimensional Empi ical Mode Decomposi ion
(BEMD) o images was in oduced in 2005 in [30].
BEMD is also a signal-dependen adap i e echnique
decomposing he image in a se ies o IMFs and
a esidue. The low-o de IMFs a e he high- equency
componen s and he high o de IMFs a e low- equency
componen s. BEMD is an adap i e mul i- esolu ion
analysis echnique ha is d i en by he inpu signal.
The e o e, i is used o a ious applica ions in image
p ocessing [31].
As he noise in image is mainly concen a ed in he
high- equency componen s, he low o de IMFs a e
ha ing mo e noise componen s as compa ed o he high
o de IMFs. Acco dingly, some BEMD based denoising
algo i hms use his ac o supp ess he noise exis ing
in low o de IMFs. Bu his may no always be ue
and a signi ican noise componen may also be p esen
in u he IMFs as well. Some denoising algo i hms
deal wi h his issue and use mu ual in o ma ion [32]
o o he il e ing echniques [33]. These echniques use
mu ual in o ma ion o a simila i y measu e be ween
he p obabili y densi y unc ions o he inpu signal and
IMFs, o de e mine he noise dominan low o de IMFs
[28], [34] and [35]. These noise-dominan IMFs a e dis-
ca ded and he signal dominan IMFs a e e ained o
ob ain he denoised signal. Howe e , mos o he high-
equency componen s o he image also con ain de ail
in o ma ion such as edges. These algo i hms he e o e
lose impo an edge de ails al hough gi ing good de-
noising pe o mance. These de ails a e o impo ance
o diagnos ic pu poses when dealing wi h ul asound
images in pa icula .
This pape in oduces an imp o ed echnique, in
which he low-o de high- equency IMFs, which
a e noise dominan , a e no comple ely disca ded.
Ra he pixel-based selec i e il e ing in o m o
Wiene il e is applied o he low o de IMFs, o
educe he noise componen , while p ese ing he edge
de ails.
The emainde o he pape is o ganized as ollows.
Sec ion 2. , p o ides a b ie o e iew o BEMD al-
go i hm along wi h pa ame e me ics used o com-
pa ison. In Sec. 3. , he p oposed denoising me hod
is desc ibed. The pe o mance e alua ion o he p o-
posed me hod is illus a ed in Sec. 4. , and Sec. 5.
p esen s he conclusions.
©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 169
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
2. Backg ound
2.1. Bidimensional Empi ical Mode
Decomposi ion
EMD in signals was in oduced by Huang e . al [26],
which is an adap i e echnique, which does no ha e
p ede ined basis unc ions and decomposes he signal
in o a ious IMFs. Simila o EMD in one dimension,
BEMD is an adap i e echnique being applied o he
images [31]. This decomposes he signal in o IMFs and
a esidue. Fo he sake o illus a ion, BEMD algo i hm
is p esen ed he e b ie ly.
Le (m, n)be he gi en image. Le l(m, n) ep-
esen he esidue o l- h IMF. To ind he nex IMF,
he esidue o he p e ious IMF is aken as he inpu .
Le il,k(m, n)be he inpu image o he gene a ion
o an IMF, whe e he i s index is l- h numbe IMF,
l= 1, . . . , L, he second index is k- h i e a ion o he
si ing p ocess, k= 1, . . . , K and (m, n)being wo spa-
ial dimensions.
S ep 1: S a wi h he gi en image as he inpu sig-
nal. i1,0(m, n) = (m, n).
S ep 2: Ex ac all local maxima and minima o
il,k(m, n).
S ep 3: In e pola e all local maxima o ge he up-
pe en elope eu(m, n)and in e pola e all local minima
o ge he lowe en elope el(m, n). The applica ion-
speci ic spline can be used o in e pola ion.
S ep 4: Calcula e he en elope mean el,k(m, n)o
he uppe and he lowe en elopes ob ained in s ep 3:
el,k(m, n) = eu(m, n) + el(m, n)
2.(1)
S ep 5: The inpu signal is upda ed by sub ac ing
he en elope mean el,k(m, n) o he nex i e a ion:
il,k(m, n) = il,k−1(m, n)−el,k(m, n), k →k+1.(2)
S ep 6: This s ep is o check i he esul ob ained in
s ep 5 is an IMF o no . Fo his s anda d de ia ion,
is calcula ed as:
=
M−1
X
m=0
N−1
X
n=0
|il,k(m, n)−il,k−1(m, n)|2
i2
l,k−1(m, n).(3)
S ep 7: Check he s anda d de ia ion  o be less
han a p ede ined alue (gene ally 0.2–0.3). I he alue
is g ea e han he c i e ion, epea s eps 2–6. When
he alue o is below he p ede ined alue, he esul
o s ep 5 is he equi ed l- h IMF, l(m, n):
l(m, n) = il,k(m, n).(4)
S ep 8: The esidue o he l- h IMF is de ined as:
l(m, n) = il,0(m, n)− l(m, n).(5)
S ep 9: The nex IMF is calcula ed by aking he
esidue calcula ed as he inpu signal and s a ing o e
om s ep 2:
il+1,0(m, n) = l(m, n).(6)
All subsequen IMFs a e calcula ed by epea ing he
s eps om 2–9. The p ocess is s opped when he
esidue calcula ed has no mo e ex ema. So, wi h he
o al Lnumbe o IMFs calcula ed and he las esidue
wi hou ex ema L, he o iginal signal can be ep e-
sen ed as:
(m, n) =
L
X
l=1
l(m, n) + L(m, n).(7)
I is impo an o men ion he e ha he low o de
IMFs, a e co esponding o he high equency while
he high o de IMFs a e co esponding o he low e-
quency.
2.2. Image Quali y Pe o mance
Me ics
The pe o mance o he p oposed BEMD based noise
educ ion echnique using a Wiene il e is analyzed
and compa ed bo h quan i a i ely and quali a i ely
wi h he exis ing echniques. Fo syn he ic US im-
ages gene a ed using Field II so wa e [36] de eloped
by J. A. Jensen, a quan i a i e analysis is ca ied ou
using pe o mance me ics iz., Peak Signal o Noise
Ra io (PSNR) [37], Edge Keeping Index (EKI) [38],
S uc u e SIMila i y Index Measu es (SSIM) [39], Co -
ela ion Coe icien (CC) [40], Signal o Noise Ra io
(SNR) [37] and Figu e o Me i (FoM) [41]. Howe e ,
o eal ul asound images, as no noise- ee image is
a ailable, he Mean o Va iance Ra io (MVR) [42] and
an Equi alen Numbe o Looks (ENL) [43] a e used o
quan i a i e e alua ion. Fo a noisy image (m, n)and
he econs uc ed denoised image (m, n), he me ics
used o quan i a i e e alua ion a e de ined as ollows:
•Peak Signal o Noise Ra io (PSNR) is a equen ly
used measu e o access he e icacy o he denoising
algo i hm. I is he a io o peak signal powe o
he noise powe gi en by Eq. (8), whe e l= 255
o an 8-bi g ayscale image o size M×N.
•Edge Keeping Index (EKI) is a pa ame e o mea-
su e he edge keeping capabili y o a denoising al-
go i hm, see Eq. (9), whe e ∆ and ∆ a e high
pass il e ed e sion o and , espec i ely, ob-
ained using 3×3Laplacian ope a o and ∆ and
170 ©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
PSNR = 10 log10
l2
1
MN
M−1
P
m=0
N−1
P
n=0
( (m, n)− (m, n))2
,(8)
EKI =
M−1
P
m=0
N−1
P
n=0 ∆ (m, n)−∆ ∆ (m, n)−∆ 
sN−1
P
n=0 ∆ (m, n)−∆ 2∆ (m, n)−∆ 2
,(9)
CC =
M−1
P
m=0
N−1
P
n=0  (m, n)−  (m, n)− 
sM−1
P
m=0
N−1
P
n=0  (m, n)− 2M−1
P
m=0
N−1
P
n=0  (m, n)− 2
.(12)
∆ a e he mean alues o ∆ and ∆ , espec-
i ely.
•S uc u al SIMila i y (SSIM) is o measu ing he
s uc u e sa ing capabili y o he denoising algo-
i hm. I and a e he mean and σiand σi
a e he s anda d de ia ion o o iginal and econ-
s uc ed images espec i ely, hen:
SSIM = 2 (m, n) (m, n) + c1(2σ , +c2)
 2+
2+c1σ2
i+σ2
, +c2,
(10)
whe e c1= (0.01l)2,c2= (0.03l)2and co a iance
image ma ix is gi en by:
σ2
, =1
N−1
N−1
X
k=0  k−  k − .(11)
•Co ela ion Coe icien (CC) de ines he in e de-
pendence o he noisy image and he econs uc ed
image. I is de ined as shown in Eq. (12).
•Signal o Noise Ra io (SNR) is de ined as he a io
o signal powe o noise powe :
SNR = 10 log10
M−1
P
m=0
N−1
P
n=0
( (m, n))2
sM−1
P
m=0
N−1
P
n=0
( (m, n)− (m, n))2
.
(13)
•Figu e o Me i (FoM) de ined by Eq. (14) is
a measu e o p ese ing he edge in o ma ion while
denoising he image:
FoM = 1
max (ND, NI)X
n=1
ND1
1 + γd2
n,(14)
whe e NDand NIa e he numbe s o edge pixels
ha a e de ec ed and ideally p esen espec i ely;
dnis he Euclidean dis ance be ween he n- h de-
ec ed edge pixel and he closes ideal pixel ha is
p esen ; γis a scala usually equal o 1
9 o image
calcula ions.
•Mean o Va iance Ra io (MVR) is calcula ed o
eal images o a selec ed local egion as:
MVR = µl
σ2
l
,(15)
whe e µland σ2
la e he local mean and a iance.
•Equi alen Numbe o Looks (ENL) is ano he pa-
ame e used o eal images. I is he a io o he
squa e o mean o he a iance o a selec ed local
egion gi en by:
ENL = µ2
l
σ2
l
.(16)
3. BEMD Based Pixel-Wise
Wiene Fil e ing
In his sec ion, he p oposed BEMD based speckle e-
duc ion in Ul asound images using Wiene il e ing
has been in oduced. Due o he high- equency cha -
ac e is ics o he speckle noise, i is mainly cons ained
in he low o de IMFs o he ul asound image. The e-
o e, conside ing he low o de IMFs o noise educ ion
is he choice ha has been conside ed.
A he same ime, he edge in o ma ion is also a piece
o c ucial in o ma ion om he poin o iew o he
diagnos ic impo ance in ul asound images. As he
edges a e signi ied by he ab up changes in he ampli-
ude in he spa ial domain, he low o de IMFs canno
be disca ded all oge he . The p oposed scheme u i-
lizes he ac ha low o de IMFs has maximum noise
©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 171
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
componen and hus low pass il e s hese IMFs using
an adap i e Wiene il e o p ese ing he edge in o -
ma ion p esen in hese low o de IMFs.
Fig. 2: Block Diag am o he p oposed scheme.
The s eps in he p oposed algo i hm a e as ollows:
S ep 1: Calcula e he BEMD o he ul asound image
co up ed wi h speckle noise.
S ep 2: Selec he low o de IMFs which ha e he
high- equency componen s and a e he e o e ha ing
bo h he noise and he edge in o ma ion.
S ep 3: Calcula e he pixel-wise Wiene il e ing
o he selec ed IMFs [44] assuming addi i e noise
(n1, n2)wi h ze o mean and a iance σ2
. Wiene il-
e es ima es he local mean and a iance a ound each
pixel o he chosen IMF:
µe=1
NM X
m,n∈w
l(m, n),(17)
σ2
e=1
NM X
m,n∈w
2
l(m, n)−µ2
e,(18)
whe e wis he window co esponding o he N×M
neighbo hood o each pixel in he IMF. This il e hen
c ea es a pixel-wise es ima e gi en as:
e(m, n) = µe+σ2
e−σ2
σ2
e
( l(m, n)−µe).(19)
S ep 4: The low-o de il e ed IMFs combined wi h
he high-o de IMFs and he esidue a e used o econ-
s uc he denoised image.
4. Expe imen al Resul s
The kidney and e us Field II images ha e been simu-
la ed and used o he expe imen s pe o med on MAT-
LAB. The speckle noise has been added o he image
wi h σ2= 0.1,σ2= 0.2, and σ2= 0.3.
Figu e 3 shows he BEMD IMFs o syn he ic kidney
image and syn he ic e us image, ha a e ob ained o
noise a iance σ2= 0.1. The spline in e pola ion ha
is used in ou case is a cubic spline. As can be seen in
Fig. 3, he low o de IMFs a e ha ing only he high-
equency componen s co esponding o speckle and
edges. The Wiene il e is bes de ined o addi i e
noise. The e o e, he mul iplica i e speckle noise has
been log- ans o med and con e ed o addi i e noise
be o e applying BEMD.
(a)
(b)
Fig. 3: BEMD IMFs o (a) Syn e ic Kidney Field II image and
(b) Syn he ic Fe us Field II image.
Pixel-wise Wiene il e ing has been applied on low
o de IMFs. I has been obse ed ha he esul s we e
bes o he Wiene il e applied o IMF1 and IMF2.
While applying Wiene il e ing, 8-neighbo hood has
been u ilized o calcula ion o local egion.
The Wiene il e ed IMFs along wi h highe -o de
IMFs and he esidue is econs uc ed o ge he de-
172 ©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING

DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
Tab. 1: Co ela ion Coe icien (CC) o a ious echniques.
Technique
CC
Noise a iance σ2
Syn he ic e us image Syn he ic kidney image
0.1 0.2 0.3 0.1 0.2 0.3
P oposed EMD 0.96234 0.94607 0.92693 0.94284 0.90968 0.87969
Con en ional EMD 0.95102 0.92555 0.87480 0.92645 0.87332 0.81060
Bila e al 0.91368 0.84291 0.78464 0.84947 0.74776 0.67000
SRAD 0.94191 0.87916 0.81739 0.91540 0.8292 0.75551
NLM 0.93126 0.87093 0.82044 0.87618 0.78484 0.71328
OBNLM 0.95155 0.90697 0.86630 0.90923 0.83734 0.77571
PMAD 0.94638 0.88481 0.82668 0.90595 0.80801 0.72333
Tab. 2: Signal o Noise Ra io (SNR) o a ious echniques.
Technique
SNR
Noise a iance σ2
Syn he ic e us image Syn he ic kidney image
0.1 0.2 0.3 0.1 0.2 0.3
P oposed EMD 14.0620 12.6740 11.4590 15.6020 13.6940 12.4430
Con en ional EMD 13.9140 11.0110 10.2900 14.2440 12.0820 10.2030
Bila e al 10.7200 7.9790 6.4897 10.7320 8.0182 6.5425
SRAD 12.6000 9.2820 7.3429 13.7580 10.2860 8.4398
NLM 11.7240 8.8585 7.3089 11.7650 8.9261 7.3852
OBNLM 13.3050 10.3730 8.6871 13.3320 10.4920 8.8753
PMAD 12.9410 9.4890 7.5779 13.2020 9.6131 7.6633
Tab. 3: Figu e o Me i (FoM) o a ious echniques.
Technique
FoM
Noise a iance σ2
Syn he ic e us image Syn he ic kidney image
0.1 0.2 0.3 0.1 0.2 0.3
P oposed EMD 0.82919 0.86146 0.84317 0.86334 0.86188 0.82560
Con en ional EMD 0.82761 0.84320 0.82133 0.86112 0.83281 0.81143
Bila e al 0.81285 0.74764 0.70383 0.77549 0.72620 0.69278
SRAD 0.86349 0.78487 0.73090 0.83815 0.75237 0.73401
NLM 0.87531 0.79219 0.75194 0.83151 0.78397 0.74317
OBNLM 0.89364 0.84390 0.80875 0.88295 0.78922 0.76806
PMAD 0.86438 0.79341 0.73724 0.85802 0.76742 0.71880
noised image. The quali a i e assessmen is shown in
Fig. 4 o syn he ic kidney and in Fig. 5 o syn he ic
e us o he noise o a iance σ2= 0.1. As can be seen,
he denoised image is pe cep ually o he same quali y
as ha o he o iginal image.
(a) (b) (c)
Fig. 4: Syn he ic kidney o σ2= 0.1: (a) o iginal image,
(b) noisy image and (c) denoised image.
Table 1, Tab. 2, Tab. 3, Tab. 4, Tab. 5 and Tab. 6
shows he compa ison o a ious pa ame e s ob ained
o di e en alues o σ o he wo Field II syn he ic
images, e us and kidney. As can be seen, he p oposed
(a) (b) (c)
Fig. 5: Syn he ic e us o σ2= 0.1: (a) o iginal image,
(b) noisy image and (c) denoised image.
algo i hm pe o ms be e han he exis ing echniques
in e ms o PSNR, SNR, and CC, o noise a iance
σ2= 0.1. The esul s ob ained o hese pa ame-
e s gi e simila esul s e en a highe noise a iances,
σ2= 0.2and σ2= 0.3.
Mo eo e , he esul in he ables show ha he p o-
posed echnique gi es compa able esul s wi h he ex-
is ing echniques in e ms o EKI, SSIM, and FoM o
©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 173
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
Tab. 4: Peak Signal o Noise Ra io (PSNR) o a ious echniques.
Technique
PSNR
Noise a iance σ2
Syn he ic e us image Syn he ic kidney image
0.1 0.2 0.3 0.1 0.2 0.3
P oposed EMD 24.946 23.577 22.383 24.322 22.388 21.158
Con en ional EMD 22.398 20.671 20.330 22.317 20.672 19.821
Bila e al 21.056 18.009 16.260 19.025 15.992 14.254
SRAD 23.058 19.463 17.260 22.192 18.459 16.398
NLM 22.336 19.310 17.594 20.288 17.239 15.518
OBNLM 24.002 21.030 19.306 21.971 19.054 17.350
PMAD 23.539 19.837 17.690 21.730 17.878 15.686
Tab. 5: Edge Keeping Index (EKI) o a ious echniques.
Technique
EKI
Noise a iance σ2
Syn he ic e us image Syn he ic kidney image
0.1 0.2 0.3 0.1 0.2 0.3
P oposed EMD 0.64603 0.59358 0.53587 0.54885 0.50024 0.49630
Con en ional EMD 0.61720 0.52081 0.50012 0.51320 0.48230 0.45321
Bila e al 0.59360 0.54006 0.52306 0.53703 0.49294 0.47574
SRAD 0.55134 0.49815 0.49498 0.49201 0.45619 0.43757
NLM 0.59768 0.54228 0.51775 0.53614 0.49024 0.47184
OBNLM 0.62562 0.54757 0.52198 0.55135 0.49023 0.46734
PMAD 0.55606 0.50568 0.49567 0.49833 0.46407 0.44813
Tab. 6: S uc u al SIMila i y (SSIM) o a ious echniques.
Technique
SSIM
Noise a iance σ2
Syn he ic e us image Syn he ic kidney image
0.1 0.2 0.3 0.1 0.2 0.3
P oposed EMD 0.70575 0.66684 0.62823 0.61088 0.53961 0.49747
Con en ional EMD 0.72345 0.65482 0.61291 0.61098 0.49308 0.45320
Bila e al 0.71808 0.61478 0.5552 0.53328 0.39544 0.32945
SRAD 0.76831 0.67000 0.59414 0.65411 0.50453 0.41941
NLM 0.74678 0.64120 0.58082 0.56396 0.42391 0.35323
OBNLM 0.78688 0.68623 0.62356 0.62079 0.47589 0.39955
PMAD 0.78014 0.68009 0.60847 0.63845 0.47608 0.38675
σ2= 0.1. A he same ime, based on he analysis o
he alues ob ained by he expe imen s, he scheme is
gi ing be e esul s in e ms o hese pa ame e s o
highe alues o σ2= 0.2and σ2= 0.3. Thus, we can
summa ize ha he echnique pe o ms accep ably well
e en a highe noise le els.
Fo comple eness o he e icacy e alua ion o he
p oposed algo i hm, an expe imen has also been pe -
o med on he eal ul asound image da abase aken
om [45]. The eal images ha e h ee se s o da a
namely kidney, li e , and gall bladde images each ha -
ing a ound 85 images. Th ee egions we e selec ed an-
domly o all h ee se s and he MVR and ENL ha e
been calcula ed.
Figu e 6(a) shows he eal li e ul asound image
and he h ee egions ha a e aken o calcula ions o
MVR and ENL. Figu e 6(b) and Fig. 6(c) a e he e-
sul s ha a e ob ained o he selec ed egions. Table 7
shows he MVR and ENL o he eal li e ul asound
da abase.
Simila expe imen s we e also pe o med on eal ul-
asound image da a se s o kidney and gall bladde ,
howe e , due o pauci y o space we ha e shown MVR
and ENL plo s o li e da abase only. The esul s ob-
ained o he o he se s a e also in conjunc ion wi h
he esul s shown he e.
Tab. 7: MVR and ENL o Real Li e Ul asound Da abase.
Technique MVR ENL
P oposed EMD 18.51 ±3.76 5.25 ±2.79
Con en ional BEMD 17.91 ±5.32 5.01 ±2.54
Bila e al 15.42 ±5.16 3.96 ±2.32
SRAD 17.66 ±4.52 4.87 ±2.35
NLM 17.01 ±4.14 4.91 ±2.15
OBNLM 17.81 ±4.71 4.95 ±2.61
PMAD 16.39 ±6.21 4.33 ±2.79
174 ©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
(a) (b) (c)
Fig. 6: (a) Real li e image wi h selec ed egions, (b) MVR plo , (c) ENL plo .
5. Conclusion
In his pape , a BEMD based image denoising algo-
i hm using pixel-wise Wiene il e ing has been p e-
sen ed. The p oposed me hod sol es he p oblem o
losing he edge in o ma ion in he con en ional BEMD
based denoising echnique. The con en ional BEMD
based algo i hm is able o emo e low o de IMF, bu
loses impo an edge in o ma ion which has been p e-
se ed wi h he help o he me hod p oposed he ein.
The pe o mance o he me hod has been e i ied quan-
i a i ely in e ms o PSNR, EKI, SSIM, FoM, SNR,
and CC o syn he ic ul asound images. The quan i a-
i e analysis is also done on he eal ul asound images
in e ms o MVR and ENL. I has been ound ha he
p oposed algo i hm pe o ms be e han many o he
exis ing s a e-o - he-a echniques and can p ese e
he edge in o ma ion.
Au ho Con ibu ions
V.K. concei ed he p esen ed idea. B.G. de eloped he
heo y and pe o med he expe imen s. V.K. encou -
aged B.G. o in es iga e and supe ised he indings
o his wo k. Bo h au ho s discussed he esul s and
con ibu ed o he inal manusc ip .
Re e ences
[1] BURCKHARDT, C. B. Speckle in ul asound
B-mode scans. IEEE T ansac ions on Sonics
and Ul asonics. 1978, ol. 25, iss. 1, pp. 1–6.
ISSN 2162-1403. DOI: 10.1109/T-SU.1978.30978.
[2] SZABO, T. L. Diagnos ic Ul asound Imaging:
Inside Ou . 2nd ed. Ams e dam: Academic P ess,
2014. ISBN 978-012-396487-8.
[3] YU, Y. and S. T. ACTON. Speckle educ-
ing aniso opic di usion. IEEE T ansac-
ions on Image P ocessing. 2002, ol. 11,
iss. 11, pp. 1260–1270. ISSN 1941-0042.
DOI: 10.1109/TIP.2002.804276.
[4] RAHIMI, M. and M. YAZDI. A new hyb id algo-
i hm o speckle noise educ ion o SAR images
based on mean-median il e and SRAD me hod.
In: 2015 2nd In e na ional Con e ence on Pa e n
Recogni ion and Image Analysis (IPRIA). Rash :
IEEE, 2015, pp. 1–6. ISBN 978-1-4799-8445-9.
DOI: 10.1109/PRIA.2015.7161623.
[5] LIU, X., J. LIU, X. XU, L. CHUN, J. TANG and
Y. DENG. A obus de ail p ese ing aniso opic
di usion o speckle educ ion in ul asound im-
ages. In: The 2010 In e na ional Con e ence on
Bioin o ma ics and Compu a ional Biology (BIO-
COMP 2010): Genomics. Las Vegas: BMC Ge-
nomics, 2011, pp. 1–10. DOI: 10.1186/1471-2164-
12-S5-S14.
[6] PERONA, P. and J. MALIK. Scale-space and
edge de ec ion using aniso opic di usion. IEEE
T ansac ions on Pa e n Analysis and Machine
In elligence. 1990, ol. 12, iss. 7, pp. 629–639.
ISSN 1939-3539. DOI: 10.1109/34.56205.
[7] TOMASI, C. and R. MANDUCHI. Bila e al il-
e ing o g ay and colo images. In: Six h In e -
na ional Con e ence on Compu e Vision. Bom-
bay: IEEE, 1998, pp. 839–846. ISBN 81-7319-221-
9. DOI: 10.1109/ICCV.1998.710815.
[8] BALOCCO, S., C. GATTA, O. PUJOL,
J. MAURI and P. RADEVA. SRBF: Speckle
Reducing Bila e al Fil e ing. Ul asound
in Medicine &Biology. 2010, ol. 36,
iss. 8, pp. 1353–1363. ISSN 1879-291X.
DOI: 10.1016/j.ul asmedbio.2010.05.007.
[9] COUPE, P., P. HELLIER, C. KERVRANN
and C. BARILLOT. Nonlocal means-based
speckle il e ing o ul asound images. IEEE
T ansac ions on Image P ocessing. 2009,
©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 175
DIGITAL IMAGE PROCESSING AND COMPUTER GRAPHICS VOLUME: 19 |NUMBER: 2 |2021 |JUNE
ol. 18, iss. 10, pp. 2221–2229. ISSN 1941-
0042. DOI: 10.1109/TIP.2009.2024064.
[10] ZHAN, Y., M. DING, L. WU and X. ZHANG.
Nonlocal means me hod using weigh e ining o
despeckling o ul asound images. Signal P ocess-
ing. 2014, ol. 103, iss. 1, pp. 201–213. ISSN 1872-
7557. DOI: 10.1016/j.sigp o.2013.12.019.
[11] LI, X., H. HE, R. WANG and J. CHENG.
Supe pixel-guided nonlocal means o image de-
noising and supe - esolu ion. Signal P ocessing.
2016, ol. 124, iss. 1, pp. 173–183. ISSN 1872-7557.
DOI: 10.1016/j.sigp o.2015.09.021.
[12] GUO, Y., Y. WANG and T. HOU. Speckle il-
e ing o ul asonic images using a modi ied non
local-based algo i hm. Biomedical Signal P ocess-
ing and Con ol. 2011, ol. 6, iss. 2, pp. 129–138.
ISSN 1746-8108. DOI: 10.1016/j.bspc.2010.10.004.
[13] KERVRANN, C., J. BOULANGER and
P. COUPE. Bayesian Non-local Means Fil e ,
Image Redundancy and Adap i e Dic iona ies
o Noise Remo al. In: SSVM: In e na ional
Con e ence on Scale Space and Va ia ional
Me hods in Compu e Vision. Ischia: Sp inge ,
2007, pp. 520–532. ISBN 978-3-540-72823-8.
DOI: 978-3-540-72823-8_45.
[14] COUPE, P., P. HELLIER, C. KERVRANN
and C. BARILLOT. Bayesian non local means-
based speckle il e ing. In: 2008 5 h IEEE
In e na ional Symposium on Biomedical Imag-
ing: F om Nano o Mac o. Pa is: IEEE,
2008, pp. 1291–1294. ISBN 978-1-4244-2002-5.
DOI: 10.1109/ISBI.2008.4541240.
[15] YANG, J., J. FAN, D. AI, X. WANG, Y. ZHENG,
S. TANG and Y. WANG. Local s a is ics and
non-local mean il e o speckle noise educ ion
in medical ul asound image. Neu ocompu ing.
2016, ol. 195, iss. 1, pp. 88–95. ISSN 1872-8286.
DOI: 10.1016/j.neucom.2015.05.140.
[16] MEI, F., D. ZHANG and Y. YANG. Imp o ed
non-local sel -simila i y measu es o e ec i e
speckle noise educ ion in ul asound images.
Compu e Me hods and P og ams in Biomedicine.
2020, ol. 196, iss. 1, pp. 1–14. ISSN 1872-7565.
DOI: 10.1016/j.cmpb.2020.105670.
[17] HUANG, S., C. TANG, M. XU, Y. QIU,
and Z. LEI. BM3D-based o al a ia ion al-
go i hm o speckle emo al wi h s uc u e-
p ese ing in OCT images. Applied Op ics. 2019,
ol. 58, iss. 23, pp. 6233–6243. ISSN 2155-3165.
DOI: 10.1364/AO.58.006233.
[18] JADWA, S. Wiene Fil e based Medical
Image De-noising. In e na ional Jou nal o
Science and Enginee ing Applica ions. 2018,
ol. 7, iss. 9, pp. 318–323. ISSN 2319-7560.
DOI: 10.7753/IJSEA0709.1014.
[19] KAY, S. M. Fundamen als o S a is ical P ocess-
ing, Volume I. 1s ed. Englewood Cli s: P en ice
Hall PTR, 1993. ISBN 978-0-1334-5711-7.
[20] BINDILATTI, A. A., M. A. C. VIEIRA and
N. D. A. MASCARENHAS. Poisson Wiene il-
e ing wi h non-local weigh ed pa ame e es ima-
ion using s ochas ic dis ances. Signal P ocessing.
2018, ol. 144, iss. 1, pp. 68–76. ISSN 1872-7557.
DOI: 10.1016/j.sigp o.2017.10.001.
[21] JAIN, P. and V. TYAGI. A su ey o edge-
p ese ing image denoising me hods. In o ma ion
Sys ems F on ie s. 2016, ol. 18, iss. 1, pp. 159–
170. ISSN 1572-9419. DOI: 10.1007/s10796-014-
9527-0.
[22] DONOHO, D. L. De-noising by so - h esholding.
IEEE T ansac ions on In o ma ion Theo y. 1995,
ol. 41, iss. 3, pp. 613–627. ISSN 1557-9654.
DOI: 10.1109/18.382009.
[23] CHOI, H. and J. JEONG. Speckle Noise Re-
duc ion Technique o SAR Images Using S a-
is ical Cha ac e is ics o Speckle Noise and Dis-
c e e Wa ele T ans o m. Remo e Sensing. 2019,
ol. 11, iss. 10, pp. 1184–1210. ISSN 2072-4292.
DOI: 10.3390/ s11101184.
[24] CHOI, H. and J. JEONG. Despeckling Algo i hm
o Remo ing Speckle Noise om Ul asound Im-
ages. Symme y. 2020, ol. 12, iss. 6, pp. 938–963.
ISSN 2073-8994. DOI: 10.3390/sym12060938.
[25] CHEN, Y., M. ZHANG, H.-M. YAN, Y.-J. LI and
K.-F. Yang. A New Ul asound Speckle Reduc-
ion Algo i hm Based on Supe pixel Segmen a-
ion and De ail Compensa ion. Applied Sciences.
2019, ol. 9, iss. 8, pp. 1693–1705. ISSN 2076-3417.
DOI: 10.3390/app9081693.
[26] HUANG, N. E., Z. SHEN, S. R. LONG,
M. C. WU, H. H. SHIH, Q. ZHENG, N.-C. YEN,
C. C. TUNG and H. H. LIU. The empi ical
mode decomposi ion and he Hilbe spec um o
nonlinea and non-s a iona y ime se ies analy-
sis. P oceedings o he Royal Socie y A: Ma he-
ma ical, Physical and Enginee ing Sciences. 1998,
ol. 454, iss. 1971, pp. 903–995. ISSN 1471-2946.
DOI: 10.1098/ spa.1998.0193.
[27] LINDERHED, A. Image Empi ical Mode De-
composi ion: a New Tool o Image P ocess-
ing. Ad ances in Adap i e Da a Analysis. 2009,
ol. 1, iss. 2, pp. 265–294. ISSN 1793-7175.
DOI: 10.1142/S1793536909000138.
176 ©2021 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING