Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
9
,
No.
6
,
Decem
ber
201
9
, p
p.
5227~
5234
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
6
.
pp5227
-
52
34
5227
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
Estimati
on of
objec
t
loc
ation pr
obability
for
object d
etectio
n
usin
g
bri
ghtness f
eatur
e on
ly
Hy
u
n
Ju
n
Par
k
1
, K
w
ang
Ba
ek Kim
2
1
Division
of
Soft
ware
Conve
rge
n
ce
,
Cheong
ju
Un
ive
rsit
y
,
Repub
lic
of
Kor
ea
2
Division
of
Co
m
pute
r
Software
Engi
n
ee
ring
,
Si
l
la
Univ
ersity
,
Re
publi
c
of
Kore
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
pr
24
, 201
9
Re
vised
Ju
l
6
,
201
9
Accepte
d
J
ul
17
, 2
01
9
Mos
t
exi
sting
ob
je
c
t
detec
t
ion
m
et
hods
use
featur
es
such
as
col
or
,
shape
,
a
nd
cont
our.
If
th
ere
are
no
consiste
nt
fea
tur
es
c
an
be
used,
we
n
ee
d
a
new
objec
t
det
e
ct
ion
m
etho
d.
Therefore,
in
thi
s
paper,
we
propose
a
new
m
et
hod
for
esti
m
at
ing
the
p
roba
bil
i
t
y
th
at
a
n
object
ca
n
b
e
loc
a
te
d
for
obj
e
ct
d
et
e
ct
ion
and
g
ene
r
at
ing
an
ob
ject
lo
ca
t
i
on
proba
b
il
i
t
y
m
ap
using
onl
y
brigh
tne
ss
in
a
gra
y
image
.
To
ev
al
ua
te
the
per
form
ance
o
f
the
proposed
m
et
hod,
w
e
appl
i
ed
it
to
ga
ll
bla
dd
er
de
tection.
Expe
r
iment
al
resul
ts
sho
wed
98.
02%
succ
ess
rate
fo
r
gallbla
dd
er
d
et
e
ct
ion
in
u
lt
r
asonogra
m
.
Th
e
ref
ore
,
th
e
proposed
m
et
ho
d
accura
t
ely
est
imate
s
th
e
obj
e
ct
lo
cation
prob
abi
lit
y
an
d
eff
ectivel
y
d
et
e
c
te
d
g
al
lb
la
dd
er.
Ke
yw
or
d
s
:
Gall
blad
der
Me
dical
i
m
age pro
ces
sin
g
Object
detect
io
n
O
bject
loc
at
ion
Ultraso
nogram
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Kw
a
ng Bae
k Kim
,
Divisio
n of Co
m
pu
te
r
Softwa
re E
ng
i
neer
i
ng,
Sil
la
U
niv
e
rsit
y,
140, Ba
egya
ng
-
dae
ro
(Blv
d)
700
be
on
-
gil
(
Rd), Sasa
ng
-
gu, B
us
a
n 469
58,
R
epubl
i
c
of
K
ore
a
.
Em
a
il
:
gb
kim
@sil
la
.ac.kr
1.
INTROD
U
CTION
Re
search
f
or
de
te
ct
ing
obj
ect
s
in
im
ages
is
on
e
of
the
m
os
t
im
po
rt
ant
fiel
ds
i
n
im
age
processi
ng.
In
ge
ne
ral,
t
he
first
thi
ng
f
or
obj
ect
detect
io
n
is
analy
zi
ng
the
featur
e
s
of
the
ob
j
ect
.
Th
eref
or
e
,
if
a
n
obj
ect
do
e
s
no
t
ha
ve
a
c
onsist
ent
siz
e,
posit
io
n,
sh
a
pe,
c
on
t
our
,
c
olor,
et
c
.,
ob
j
e
ct
de
te
ct
ion
beco
m
es
di
ff
ic
ult.
In
pa
rtic
ular,
wh
e
n
t
he
relat
ive
bri
ghtnes
s
is
the
on
ly
usa
ble
feat
ur
e
,
an
a
dd
it
io
nal
pro
blem
of
thr
esh
old
set
ti
ng
is
occ
urred, a
nd the
de
te
ct
ion
r
es
ult i
s
d
if
fer
e
nt
dep
e
nd
i
ng on t
he us
ed
th
res
ho
l
d va
lue [1,
2
]
.
Since
m
os
t
exi
sti
ng
ob
j
ect
de
te
ct
ion
m
et
ho
ds
us
e
feat
ur
es
s
uch
as
c
olor,
s
hap
e
,
an
d
co
nt
our,
if
the
re
are
no
c
onsist
ent
feat
ur
es
i
n
a
gr
ay
im
age
or
the
obj
ect
ca
n
be
detect
ed
by
on
ly
the
dif
f
eren
ce
in
br
i
ghtness
with
the
s
urroundin
gs,
we
ne
ed
a
ne
w
ob
je
ct
detect
ion
m
et
ho
d.
The
re
fore,
in
t
his
pa
per,
we
pro
po
s
e
a
new
m
et
ho
d
to
e
sti
m
at
e
the
pro
ba
bili
ty
that
an
obj
ect
can
be
lo
cat
ed
in
a
n
im
age
us
in
g
only
bri
ghtness
fea
ture
,
and
ex
plain
a
m
e
tho
d
of
ge
ner
at
in
g
a
n
obj
ect
lo
cat
io
n
pro
bab
il
it
y
m
ap
that
e
xpr
esses
the
est
im
at
ed
pro
bab
il
it
y as
an
im
age.
In
orde
r
to
est
i
m
at
e
the
obj
ec
t
locat
ion
probabil
it
y,
we
en
ha
nce
the
c
ontra
st
us
in
g
the
e
nds
-
i
n
searc
h
stret
chin
g
[
3]
and
us
e
the
co
lor
qua
ntiza
ti
on
m
et
ho
d
us
in
g
c
olor
im
po
rtance
-
base
d
sel
f
-
orga
nizing
m
ap
[
4]
.
The
pro
po
se
d
m
et
ho
d
ca
n
est
i
m
at
e
the
locat
ion
of
an
obj
ec
t
hav
e
no
c
on
si
ste
nt
featu
res
i
n
gray
im
age
even
it
cannot
be
dete
ct
ed
by
us
i
ng
ex
ist
ing
ob
j
ect
detect
io
n
m
eth
ods.
I
n
ad
diti
on,
si
nce
the
re
is
no
pa
ram
eter
or
thres
ho
l
d
set
ti
ng,
c
onsta
nt
re
su
lt
is
ge
nerat
ed
a
nd
co
ns
ta
nt
execu
ti
on
ti
m
e
is
re
qu
ire
d.
T
her
e
fore,
it
is
s
uitable
for real
-
ti
m
e i
m
age pro
ces
sing.
Ultraso
nogram
s
are
gr
ay
im
a
ges
a
nd
the
re
i
s
no
c
olor
in
form
ation
.
Be
ca
use
of
the
featu
r
es
of
organs
in
ultras
onogra
m
are
ver
y
dif
f
eren
t
dep
e
ndin
g
on
t
he
pr
of
ic
ie
ncy
of
the
operator
[
5],
the
orga
ns
a
pp
ea
ring
i
n
ultraso
nogram
s
are
s
hould
be
detect
ed
on
l
y
by
the
dif
fere
nce
in
br
i
gh
t
ness
[
6]
.
Pa
rtic
ularly
,
the
gallbla
dde
r
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5227
-
5234
5228
that
app
ea
rs
in
ultraso
nogram
is
an
obje
ct
w
it
h
no
c
onsist
ent
featu
res
[
7
-
10
]
.
T
he
refor
e
,
it
is
a
go
od
fi
el
d
to
app
ly
the
pro
po
s
ed
obj
ect
l
ocati
on
pro
ba
bili
ty
m
ap,
so
the
pro
posed
m
et
ho
d
will
be
ex
plaine
d
us
in
g
ultraso
nogram
s which
is co
nt
ai
nin
g gall
blad
der.
A
relat
ed
s
tud
ie
s ca
n be
f
ound in
Refe
re
nce
[11
-
14]
.
2.
OBJECT
LO
CA
TI
ON PR
OB
ABILIT
Y
MAP GE
NER
ATI
O
N MET
HOD
Figure
1
show
s
the
proces
s
of
est
im
a
ti
ng
obj
ect
l
ocati
on
pro
ba
bili
ty
and
gen
e
rati
ng
a
pro
ba
bili
ty
m
ap.
W
e
est
im
at
e
t
he
area
wh
e
re
t
he
obje
ct
can
be
l
oc
at
ed,
a
nd
ge
ne
rate
10
0
m
ulti
-
locat
io
n
in
f
orm
at
ion
.
The
obj
ect
l
oc
at
ion
pro
bab
il
it
y
is
est
i
m
a
te
d
base
d
on
t
he
m
ulti
-
locat
ion
in
form
ation
,
a
nd
the
pro
bab
il
it
y
m
ap
is ge
ner
at
ed
.
Figure
1. The
process
of esti
m
at
ing
obje
ct
locati
on pr
ob
a
bi
li
t
y and
ge
nerat
ing
a
pro
ba
bili
ty
m
ap
2
.
1.
Prepr
oce
ssing
The
pr
e
process
ing
us
es
bilat
eral
filt
er
[
15,
16]
an
d
m
edian
filt
er
[17,
18]
to
rem
ov
e
noise
.
Fig
ur
e
2
sh
ows
the
res
ult
of
a
pply
ing
bilat
eral
filt
er
a
nd
m
edian
filt
er.
The
bi
l
at
eral
filt
er
is
non
-
li
near
an
d
is
wel
l
known
as
a
no
ise
-
re
duci
ng
sm
oo
thin
g
fi
lt
er
that
rem
ov
es
noise
w
hi
le
pr
ese
r
ving
edg
e
of
a
n
i
m
age
.
The
m
edian
filt
er
ha
s
t
he
e
ffec
t
of
rem
ov
in
g
t
he
im
pu
lse
no
ise
w
hile
al
so
preser
ving
t
he
e
dge.
The
r
efore,
by u
si
ng the
ab
ov
e
tw
o fil
te
rs, it
is possible t
o pr
ese
r
ve
the
edg
e
and
rem
ov
e
no
ise
s
.
(a)
(b)
Figure
2
.
Edge
-
preser
ving
noise
r
em
ov
al
usi
ng b
il
at
eral
filt
er a
nd m
edian
filt
er
,
(a)
O
rigin
al
im
age
,
(
b)
N
oise
rem
ov
ed
im
age
2.2.
O
b
ject
lo
cat
i
on
proba
b
il
ity
represe
ntati
on
u
sing
en
ds
-
in se
arc
h stretchi
n
g
In
this
pa
per
,
we
assum
e
t
he
ta
rget
obj
e
ct
’s
br
i
gh
t
ness
is
l
ow
er
tha
n
the
s
urrou
ndin
g
are
a
.
If
the
br
i
gh
t
ne
ss
of
the
obj
ec
t
is
higher
t
ha
n
t
he
surr
ound
ing
area
,
set
it
to
the
opposit
e
of
the
descr
i
ption
belo
w.
Fig
ure
3
s
how
s
how
to
represe
nt
t
he
ob
j
ect
locat
io
n
pro
bab
il
it
y
by
br
i
gh
t
ness.
The
ob
j
ect
loc
at
io
n
pro
bab
il
it
y
is
def
i
ned
a
s
the
pro
bab
il
it
y
that
the
obj
ect
ca
n
be
locat
ed
.
Sinc
e
the
obj
e
ct
is
assum
ed
to
ha
ve
a
lo
w
bri
ghtne
ss,
i
f
t
he
bri
gh
tness
T
low
incl
ud
e
d
in
the
bri
gh
t
ness
ra
nge
of
the
obj
ect
,
t
he
ra
ng
e
belo
w
T
low
beco
m
es
a
ra
nge
in
w
hic
h
t
he
obj
ect
ca
n
be
locat
e
d
a
nd
def
i
nes
it
as
a
m
ini
m
u
m
l
ocatable
range
R
min
.
The
obj
ect
ca
n
be
l
ocated
i
n
R
min
with
a
high
pro
bab
il
it
y.
Con
tra
sti
vely
,
if
we
kn
ow
t
he
m
ini
m
u
m
br
ig
htn
es
s
Thigh
that
t
he
obj
ect
ca
nnot
be
locat
e
d,
t
he
n
a
br
i
gh
t
nes
s
gr
e
at
er
tha
n
T
high
is
a
range
in
w
hich
the
obj
ect
nev
e
r
ca
n
be
locat
ed.
The
re
f
or
e
,
the
range
exclu
ding
gr
ea
te
r
than
Thi
gh
is
def
i
ned
a
s
th
e
m
axi
m
u
m
locat
able
range
R
max
.
Sin
ce
R
m
ax
has
h
i
gher
brig
htn
ess
than
R
min
,
R
max
has
a
r
el
at
ively
low ob
j
ect
loc
at
ion
pro
bab
il
it
y.
B
i
l
a
t
e
r
a
l
f
i
l
t
e
r
M
e
d
i
a
n
f
i
l
t
e
r
C
h
a
n
g
e
p
a
r
a
m
e
t
e
r
s
o
f
e
n
d
s
-
i
n
s
e
a
r
c
h
s
t
r
e
t
c
h
i
n
g
E
n
d
s
-
i
n
s
e
a
r
c
h
s
t
r
e
t
c
h
i
n
g
Q
u
a
n
t
i
z
a
t
i
o
n
G
e
n
e
r
a
t
e
a
l
o
c
a
t
a
b
l
e
i
m
a
g
e
20
l
o
c
a
t
a
b
l
e
i
m
a
g
e
s
?
G
e
n
e
r
a
t
e
m
u
l
t
i
p
l
e
-
l
o
c
a
t
i
o
n
i
n
f
o
r
m
a
t
i
o
n
No
Y
e
s
O
r
i
g
i
n
a
l
i
m
a
g
e
P
r
o
b
a
b
i
l
i
t
y
m
a
p
G
e
n
e
r
a
t
e
o
b
j
e
c
t
l
o
c
a
t
i
o
n
p
r
o
b
a
b
i
l
i
t
y
a
n
d
m
a
p
2
.
1
2
.
2
2
.
4
2
.
5
2
.
3
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Esti
ma
ti
on
of object
loc
ation
pr
ob
ab
il
it
y for
ob
je
ct
detect
io
n usin
g brig
htne
ss featu
re
on
l
y
(
Hyun
J
un P
ar
k
)
5229
Figure
3
.
Th
e
re
la
ti
on
sh
i
p between
brig
htn
es
s and
obj
ect
l
oc
at
ion
pro
bab
il
it
y
En
ds
-
i
n
searc
h
stret
chin
g
[3
]
is
on
e
of
the
c
on
t
rast
stret
chi
ng
m
et
ho
ds
in
wh
ic
h
a
ce
rtai
n
am
ou
nt
of
pix
el
s
is
set
to
wh
it
e
or
bl
ack
a
nd
the
r
e
m
ai
nin
g
pix
e
ls
are
stret
c
he
d
to
a
value
betwee
n
0
a
nd
255.
The
(1)
e
xpla
ins t
he
e
nd
s
-
in
search
stretchi
ng.
)
,
(
,
)
,
(
,
)
,
(
,
255
)
/(
)
)
,
(
(
255
0
))
,
(
(
y
x
i
h
i
g
h
f
o
r
h
i
g
h
y
x
i
l
o
w
f
o
r
l
o
w
y
x
i
f
o
r
l
o
w
h
i
g
h
l
o
w
y
x
i
y
x
i
o
u
t
p
u
t
(1)
wh
e
re
i
(
x
,
y
)
i
s
the
br
i
gh
t
nes
s
of
the
x
,
y
c
oor
din
at
e
pix
el
of
the
im
age,
l
ow
is
a
t
hr
e
shold
f
or
set
ti
ng
pix
el
s
with
a
l
ow
e
r
br
i
gh
t
ness
th
a
n
low
t
o
black
(
0),
a
nd
hi
gh
is
a
t
hr
es
ho
ld
f
or
set
ti
ng
pix
el
s
with
a
higher
br
i
gh
t
ness
tha
n
hi
gh
to
wh
it
e
(
255).
If
low
=
50
an
d
high
=
150,
pixe
ls
wit
h
a
br
i
ghtness
lo
wer
t
ha
n
50
in
the
i
m
age
are
set
to
black,
a
nd
pix
el
s
with
a
br
ig
htn
es
s
higher
tha
n
150
are
set
to
w
hite.
A
nd
pix
e
ls
wit
h
the brig
htn
es
s
betwee
n 51 an
d 149 is
stretch
ed
to
b
et
ween
0
a
nd
25
5.
If
the
par
am
et
er
of
en
ds
-
i
n
se
arch
stret
c
hing
low
is
se
t
to
T
low
and
hi
gh
is
set
to
T
high
,
we
can
obta
in
the
re
su
lt
that
the
bri
ghtnes
s
bel
ow
T
low
i
s
blac
k,
the
bri
ghtness
ab
ov
e
T
high
is
wh
it
e,
a
nd
the
bri
gh
t
ness
betwee
n
T
low
a
nd
T
high
is
0
t
o
255.
Sin
ce
we
assum
ed
that
the
ob
j
ect
’s
bri
gh
tne
ss
is
lo
w
,
the
obj
e
ct
l
oc
at
ion
pro
bab
il
it
y
increases
as
the
br
i
gh
t
ness
dec
reases,
an
d
de
creases
as
the
bri
ghtness
in
creases.
The
re
fore
,
by
en
han
ci
ng
the
co
ntrast
by
us
i
ng
e
nds
-
in
sea
rch
st
r
et
ching,
the
obj
ect
lo
cat
ion
pro
bab
il
it
y
can
be
appr
opriat
el
y expresse
d
by th
e
bri
ghtn
ess
.
2.3. De
termin
ing
ob
j
ect l
oc
atab
le
r
ange
u
sing
c
olo
r
qua
nt
iz
at
ion
Since
t
he
c
on
t
r
ast
en
han
ce
d
im
age
sti
ll
has
a
bri
ghtness
be
tween
0
an
d
255,
th
ere
is
a
pro
blem
that
the
br
i
gh
t
ness
ra
nge
i
n
w
hich
t
he
ob
j
ect
can
be
locat
ed
finall
y
is
determ
ined.
T
he
pro
posed
m
et
ho
d
determ
ines
the
bri
ghtness
ra
nge
by
usi
ng
c
ol
or
quantiz
at
io
n
m
et
ho
d.
Col
or
qua
ntiza
ti
on
[
19
-
25]
is
a
m
et
hod
to
re
p
resen
t
th
e
col
or
s
of orig
inal
im
age
with
li
m
i
te
d
num
ber
o
f
col
or
s
. I
t cl
us
te
rs
or
le
ar
ns
t
he
or
i
gin
al
co
lo
rs
and
c
onve
rts
t
he
c
olors
with
in
a
certai
n
ra
ng
e
int
o
opti
m
iz
ed
re
pr
ese
ntati
ve
c
olor.
That
is,
by
a
pply
ing
the
qua
ntiza
ti
on
,
t
he
br
i
gh
t
ne
ss
of o
ri
gin
al
i
m
age
is
di
vid
e
d
int
o
s
pecific ranges,
an
d
t
he
pix
el
s
co
rr
es
pondin
g
to each
r
a
nge a
re c
onver
te
d
i
nt
o
re
pr
e
sentat
ive c
olors.
In
the
propose
d
m
et
ho
d,
we
qu
a
ntize
t
he
c
on
t
rast
e
nhanc
ed
im
age
with
8
c
olors
(C
1
~
C
8
)
by
us
i
ng
colo
r
qu
a
ntiza
ti
on
m
et
ho
d
us
in
g
c
olor
im
po
rtance
-
bas
ed
sel
f
-
orga
nizing
m
aps
[
4]
.
Fig
ur
e
4
s
hows
the
pr
ogress
of
determ
ining
the
ob
j
ect
loc
at
able
ra
ng
e
s.
Since
we
as
sum
ed
the
obj
ec
t
is
da
rk,
it
is
al
s
o
expresse
d
i
n
dark
c
olor
i
n
the
quantiz
at
ion
res
ult.
T
he
refor
e
,
the
bri
ghtness
ra
ng
e
of
t
he
tw
o
darkest
represe
ntati
ve
colo
rs
(C
1
,
C
2
)
is
a
hi
gh
prob
a
bili
ty
that
the
obj
ect
can
be
l
ocated,
a
nd
this
is
de
fi
ned
as
the m
ini
m
u
m
o
bj
ect
loc
at
able
range
LR
min
. O
n
the c
on
t
rar
y,
t
he
bri
ghtness
range e
xcep
t f
or
t
he
bri
ghte
st colo
r
(C
8
)
is
a
relat
ively
low
pr
obabili
ty
that
the
ob
j
ect
can
be
locat
e
d,
a
nd
is
de
fine
d
a
s
a
m
axi
m
u
m
obj
ect
locat
able ra
nge
LR
max
.
0
255
T
l
o
w
=
low
T
h
i
g
h
=
high
Not loca
table
ra
ng
e
Min.
loca
table
ra
ng
e
R
m
i
n
Or
ig
inal
R
e
s
u
l
t
o
f
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t
0
0~
255
255
Ma
x
.
loc
a
ta
ble
r
a
ng
e
R
m
a
x
Obje
c
t
ra
ng
e
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5227
-
5234
5230
Figure
4
.
D
et
er
m
ining
the
br
i
gh
t
ness ra
ng
e
of an
ob
j
ect
lo
cat
able re
gion
us
in
g q
uan
ti
zat
ion
2.4. Estim
at
i
ng
m
ultipl
e
-
l
oc
at
i
on
in
fo
r
m
at
i
on
The
locat
io
n
in
form
ation
m
ea
ns
an
im
age
in
wh
ic
h
the
obj
e
ct
locat
able
re
gions
are
m
arked,
a
nd
it
is
est
i
m
at
ed
by
app
ly
in
g
AND
op
e
rati
on
to
t
he
m
ini
m
u
m
locat
able
i
m
age
LI
min
an
d
the
m
axi
m
u
m
locat
able
i
m
age
LI
max
.
LI
min
is
an
im
age
obta
ined
by
conve
rting
pi
xe
ls
inclu
ded
i
n
LR
min
into
bla
ck
(
0)
a
nd
t
he
rest
pix
el
s a
re c
onve
rted
i
nto
w
hite (
255), a
nd
LI
max
is an im
age co
nve
rted
as
a
bove usi
ng
LR
max
.
The
pro
posed
m
et
ho
d
ge
ner
a
te
s
LI
min
an
d
LI
max
re
peatedly
(20
ti
m
es)
whil
e
c
hangin
g
t
he
lo
w
a
nd
high
values
of
the
e
nds
-
i
n
se
arch
stret
chi
ng
pa
ram
et
ers.
Since
it
is
im
po
ssible
to
kn
ow
the
e
xact
valu
es
of
T
low
and
T
high
for
est
im
a
ti
ng
the
ob
j
ect
'
s
br
ig
htn
ess
ra
nge,
it
is
intende
d
to
est
im
at
e
the
ob
j
ect
lo
cat
io
n
pro
bab
il
it
y
in
var
i
o
us
bri
ghtness
c
onditi
on
s
by
ass
um
ing
T
low
and
T
high
as
var
i
ous
val
ue
s.
Fig
ur
e
5
s
ho
ws
a
n
exam
ple o
f
LI
m
in
and
LI
max
. T
he
p
a
ram
et
er v
al
ues used t
o ge
ner
at
e
LI
min
a
nd
LI
max
are
sho
wn in Fi
gure
5.
Figure
5. Mi
nim
u
m
an
d
m
axi
m
u
m
locat
able i
m
ages,
(a)
Mi
nim
u
m
l
ocatable i
m
ages
,
(
b) Maxim
um
locat
able im
ages
AND
op
e
rati
on
ext
racts
only
the
re
gions
in
dicat
ed
by
bla
ck
in
LI
min
an
d
LI
max
.
I
n
the
t
wo
im
ages,
black
m
eans
that
the
obj
ect
locat
able
reg
i
on,
so
it
is
t
he
sam
e
as
extrac
ti
ng
t
he
reg
i
ons
w
he
re
in
dica
te
d
as
obj
ect
locat
abl
e.
Be
cause
LI
min
and
LI
max
are
gen
e
rated
us
i
ng
t
he
assum
ed
T
low
an
d
T
high
values,
s
o
it
m
ay
be
Q
u
a
n
t
i
z
a
t
i
o
n
r
e
s
u
l
t
C
1
C
2
C
8
…
R
e
p
r
e
s
e
n
t
a
t
i
v
e
ER
m
i
n
0
255
T
low
=
low
T
high
=
high
R
e
s
u
l
t
o
f
c
o
n
t
r
a
s
t
e
n
h
a
n
c
e
m
e
n
t
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Esti
ma
ti
on
of object
loc
ation
pr
ob
ab
il
it
y for
ob
je
ct
detect
io
n usin
g brig
htne
ss featu
re
on
l
y
(
Hyun
J
un P
ar
k
)
5231
inaccu
rate,
A
ND
op
e
rati
on
extracts
t
he
re
gion
w
her
e
the
ob
j
ect
lo
cat
io
n
pro
ba
bili
ty
i
s
hi
gher
by
ex
tract
in
g
the
com
m
on
r
egio
ns
.
I
n
a
ddit
ion
,
the
com
bin
at
io
n
of
LI
min
and
LI
max
i
s
inten
ded
to
est
i
m
at
e
m
or
e
locat
ion
inf
or
m
at
ion
in
m
or
e
var
i
ou
s
cases
a
nd
to
est
im
a
te
m
or
e
acc
ur
at
e
ob
je
ct
l
ocati
on
pro
ba
bili
ty
ther
eafter.
In
the
e
stim
at
e
d
locat
io
n
in
form
ation
,
the
re
gions
to
o
bi
g
or
to
o
sm
al
l
ar
e
rem
ov
ed
usi
ng
siz
e
filt
er
be
cause
the
obj
ect
ha
s
a
c
ertai
n
si
ze.
Fi
gure
6
sh
ows
the
process
of
e
sti
m
at
ing
a
l
oca
ti
on
i
nfor
m
at
i
on
o
f
the g
al
lbla
dder
in ult
rason
ogr
a
m
.
(a)
(b)
(c)
(d)
Figure
6
.
Th
e
process
of esti
m
at
ing
a locati
on in
form
at
ion
,
(a)
LI
min
, (
b)
LI
max
, (
c)
AND o
per
at
io
n,
(
d) L
ocati
on in
form
at
ion
(size fil
te
r
is a
pp
li
ed
)
The
pr
opos
e
d
m
et
ho
d
est
im
a
te
s
a
total
of
100
pieces
of
posit
ion
i
nfor
m
at
ion
by
a
pp
ly
ing
a
n
A
ND
op
e
rati
on
to
e
ach
of
10
LI
min
and
LI
max
al
te
rn
at
el
y,
a
nd
def
i
nes
it
as
m
ul
ti
-
locat
ion
inf
or
m
at
ion
.
F
igure
7
sh
ows
25
of
t
he
total
100
m
ul
ti
-
locat
ion
inf
or
m
at
ion
im
ages
f
or
det
ect
ing
gallbla
dder
i
n
ultraso
nogram
.
The
act
ual
gall
blad
der
reg
i
on
is
i
nclu
de
d
i
n
21
of
25
im
a
ges.
Howe
ver
,
the
non
-
gallbl
add
e
r
re
gions
can
be
seen
only
in
s
om
e i
m
ages.
Figure
7. Mult
iple
-
locat
io
n
i
nfor
m
at
ion
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5227
-
5234
5232
2.5. Ob
ject
lo
cat
i
on
proba
b
il
ity
estim
at
i
on a
nd pro
babi
li
ty
m
ap
gene
rat
i
on
By
aver
agi
ng
the
est
i
m
at
ed
m
ulti
ple
-
locat
ion
in
form
at
i
on,
it
is
po
ssi
ble
to
in
direct
ly
est
i
m
at
e
the
pro
ba
bili
ty
that
an
obj
ec
t
can
be
locat
ed
at
eac
h
pi
xe
l
po
sit
io
n
us
ing
bri
ghtness
.
This
pro
bab
il
it
y
is
def
i
ned
as
the
obj
ect
l
ocati
on
pro
bab
il
it
y,
and
the
im
age
re
pr
ese
ntati
on
is
de
fine
d
as
the
pr
ob
a
bili
ty
m
ap.
Figure
8
sho
w
s
an
exam
ple
of
ob
j
ect
locat
ion
pro
ba
bili
ty
m
ap
for
detect
ing
gallbla
dde
r
in
ultras
onog
ram
.
In
ob
j
ect
locat
ion
pro
ba
bili
ty
m
ap,
the
da
r
k
reg
i
ons
in
di
cat
e
that
the
obj
ect
ca
n
be
hi
gh
ly
lo
cat
ed
,
an
d
the act
ual
gallb
la
dd
e
r
re
gion
a
pp
ea
rs da
rk
e
r
t
han o
t
her re
gions.
Figure
8
.
O
bje
ct
l
ocati
on
pro
bab
il
it
y m
ap
3.
E
X
PERI
MEN
T
AL
RES
UL
TS A
ND AN
A
LYSIS
The
e
xperim
e
ntal
en
vir
on
m
ent
f
or
eval
uating
t
he
pro
po
s
ed
obj
ect
loc
a
ti
on
pro
bab
il
it
y
m
ap
is
as
fo
ll
ows.
W
e
use
d
PC
with
I
nt
el
i5
-
44
60
3.2
0
GH
z
CPU
an
d
8.0
GB
RAM,
an
d
the
m
et
hod
is
im
ple
m
ent
ed
in
Visu
al
Stu
dio
2015
with
MF
C
-
base
d
C
++
la
nguag
e
.
O
pe
nCV
3.1
0
was
us
e
d
as
a
n
im
a
ge
processi
ng
l
ibrar
y.
The
98
ultraso
nogram
s
us
ed
i
n
the
e
xp
e
rim
e
nt
we
re
ta
ke
n
us
in
g
PHILI
PS
I
U
22
Ultras
ound
e
quipm
ent
from
Novem
ber
2013 to
J
uly 2
016. All ult
ras
onog
ram
s ar
e u
se
d
i
n
the
te
rtia
ry
hosp
it
al
.
3.1.
Gr
ou
n
d
t
ruth
f
or test
ima
ges
To
eval
uate
t
he
pe
rfo
rm
ance
of
the
pro
po
se
d
m
et
hod,
gro
und
tr
uth
s
wer
e
gen
e
rated
for
t
he
c
ollec
te
d
ultraso
nogram
s.
T
he
gro
und
truth
is
c
om
po
sed
of
the
cent
er
po
i
nt
of
t
he
gallbla
dder
,
th
e
le
ng
t
h
of
the
m
ajor
axis
a
nd
t
he
le
ng
t
h
of
t
he
m
i
nor
axis,
a
nd
t
he
a
rea
of
the
gallbla
dder
.
Ta
ble
1
sho
ws
ex
a
m
ples
of
ge
ne
rated
gro
und
trut
h.
The
gal
lbla
dd
e
r
a
rea
was
dire
ct
ly
m
ark
ed
by
the
pe
rs
on,
an
d
t
he
ce
nter
point,
le
ngt
h,
an
d
are
a
of
t
he
gallbla
dder
wer
e
obta
ined
us
i
ng
t
he
pro
gr
am
.
If
the
gallbla
dder
lo
ok
s
li
ke
se
par
a
te
d
[
26,
27]
as
sh
ow
n
in
Table
1
(
b),
the
area
is
c
al
culat
ed
as
th
e
su
m
of
the
t
wo
ar
eas,
a
nd
the
ce
nter
point
an
d
le
ngt
h
are
cal
culat
ed
as t
he
a
ver
a
ge of
t
he
tw
o
a
rea.
Table
1
.
E
xam
ples
of
gro
und
truth
(
un
it
:
pixe
l
)
(a)
(b)
(c)
Grou
n
d
tr
u
th
Cen
ter
p
o
in
t (
x
,
y
)
(30
6
,
2
2
6
)
(16
7
,
2
0
8
)
(19
8
,
1
7
8
)
L
en
g
th
(
m
ajo
r
,
m
i
n
o
r)
(15
7
,
8
2
)
(16
6
,
4
9
)
(33
3
,
1
3
1
)
A
rea
9
,59
8
1
2
,81
5
3
1
,37
0
3.2.
Perf
orm
ance
evalu
at
i
on o
f obj
ec
t
l
ocation
pr
obabil
ity
We
ex
per
im
ented
how
m
any
the
act
ual
obj
e
ct
reg
io
ns
is
in
cl
ud
e
d
in
the
ge
ner
at
e
d
pro
ba
bili
ty
m
ap
to
eval
uate
the
perform
ance
of
t
he
est
im
a
ted
ob
j
ect
locat
ion
pro
ba
bili
ty
.
To
do
t
his,
th
e
pro
bab
il
it
y
m
ap
is
bin
a
rized w
it
h t
he
th
res
ho
l
d
of 8
5
(=
25
5
×
1/
3)
t
o
gen
e
rate a
can
did
at
e
im
age.
If
the
can
did
at
es
c
onta
in
m
or
e
than
70%
of
t
he
gallbla
dd
e
r
r
egio
n
i
n
gro
und
tr
uth,
it
is
ju
dg
e
d
that
the
obj
ect
locat
io
n
pro
bab
il
it
y
est
im
at
ion
is success
f
ul.
T
able 2 s
hows
t
he result
of ob
je
ct
cand
i
date
ge
ner
at
io
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Esti
ma
ti
on
of object
loc
ation
pr
ob
ab
il
it
y for
ob
je
ct
detect
io
n usin
g brig
htne
ss featu
re
on
l
y
(
Hyun
J
un P
ar
k
)
523
3
Table
2
. Res
ult o
f
ca
nd
i
date ge
ner
at
io
n usi
ng obj
ect
l
ocati
on
pro
bab
il
it
y
Res
u
lt
Total cand
id
ates
270
#
of
ob
ject
101
Hit r
atio
3
6
.67
% (99
/2
7
0
)
Esti
m
atio
n
su
ccess
r
ate
9
8
.02
% (99
/1
0
1
)
A
total
of
270
cand
i
dates
we
re
ge
ner
at
e
d
f
r
om
the
cand
id
at
e
i
m
ages.
Since
ther
e
are
3
im
ages
of
gallbla
dder
d
i
vi
ded
i
nto
t
wo in th
e
98 e
xperi
m
ent i
m
ages,
there
are
101 ac
tual o
bj
ect
s
. T
he
re
su
lt
s
how
s 99 of
the
10
1
obj
ect
wer
e
inclu
de
d
in
the
c
an
did
at
es,
w
hic
h
s
how
ed
98.
02%
obj
ect
su
cces
s
r
at
e,
a
nd
99
of
t
he
27
0
cand
i
dates a
re
est
i
m
at
ed
as act
ual obj
e
ct
s,
a
nd the
h
it
rate i
s
36.67%
.
In
a
ddit
ion
,
w
e
cal
culat
ed
th
e
differe
nce
in
br
i
gh
t
ness
between
t
he
obj
e
ct
(g
al
lblad
de
r
)
re
gion
a
nd
the
non
-
ob
j
ect
(
non
-
ga
ll
blad
der)
re
gion
in
the
gen
e
rated
pr
ob
a
bili
ty
m
ap.
T
he
ob
j
ect
re
gion
is
bas
ed
on
gro
und
t
ru
t
h.
Table
3
s
how
s
the
ave
rag
e
bri
ghtness
of
the
obj
ect
re
gi
on
a
nd
the
non
-
ob
j
ect
re
gi
on
i
n
the pr
ob
a
bili
ty
m
ap.
Table
3
. A
ve
ra
ge bri
gh
t
ness o
f object
a
nd no
n
-
obj
ect
reg
i
on
s in pro
bab
il
it
y m
ap
Av
erage brig
h
tn
ess
of
o
b
ject r
eg
io
n
Av
erage brig
h
tn
ess
of
n
o
n
-
o
b
ject r
eg
io
n
M
in
i
m
u
m
1
5
.66
1
9
7
.65
M
ed
ian
7
3
.71
2
2
8
.86
M
ax
i
m
u
m
2
4
8
.47
(t
op
1
0
% = 13
6
.88
)
2
5
0
.71
(
t
op
1
0
% = 24
3
.75
)
A
v
erage
7
9
.77
2
2
7
.38
In
a
pro
bab
il
it
y
m
ap,
br
i
gh
t
ne
ss
m
eans
the
obj
ect
l
ocati
on
pro
ba
bili
ty
,
and
the
lo
wer
th
e
bri
ghtness
,
the
highe
r
the
pro
bab
il
it
y.
Th
e
aver
a
ge
br
i
ghtness
of
the
obj
ect
reg
i
on
is
about
35%
of
the
ave
ra
ge
bri
gh
t
ness
of
the
no
n
-
obj
ect
re
gion
,
s
o
that
it
c
onfir
m
ed
that
th
e
est
i
m
at
ed
obj
e
ct
locat
ion
prob
a
bili
ty
is
hi
gh
in
the
re
gion
w
he
re
the
act
ua
l
obj
ect
is
loc
at
ed.
Wh
e
n
t
he
est
im
a
te
d
obj
ect
locat
io
n
is
not
accu
ratel
y,
the
a
ver
a
ge
br
igh
tne
ss
of
t
he
ob
j
ect
reg
i
on
inc
reases
up
t
o
248.4
7.
I
t
is
a
c
ounte
rev
i
de
nce
t
hat
the
obj
ect
locat
ion
pro
ba
bili
ty
is accur
a
te
ly
estim
at
ed.
4.
CONCL
US
I
O
N
In
this
pa
per
,
we
pro
pose
d
a
new
m
et
ho
d
for
est
i
m
at
ing
the
prob
a
bili
ty
t
hat
an
obj
ect
c
an
be
l
ocate
d
for
ob
j
ect
dete
ct
ion
a
nd
ge
ne
rati
ng
a
n
obj
ec
t
locat
ion
pro
ba
bili
ty
m
ap
us
ing
only
bri
ghtness
in
a
gray
i
m
age
wh
e
re
i
nfor
m
at
ion
s
uc
h
as
c
olor,
sh
a
pe,
a
nd
co
nt
our
ca
nnot
be
us
e
d.
The
propose
d
m
et
ho
d
repre
sented
the
pro
bab
il
it
y
that
an
obj
ect
ca
n
be
locat
e
d
us
i
ng
th
e
bri
gh
tne
ss
us
i
ng
the
e
nds
-
in
s
earch
stret
c
hing
a
nd
qu
a
ntiza
ti
on,
and
gen
e
rate
d
t
he
m
ulti
-
l
ocati
on
in
f
orm
at
ion
.
Ba
se
d
on
m
ulti
-
l
ocati
on
i
nf
orm
at
ion
,
the
pr
opos
e
d
m
et
ho
d
e
stim
a
te
d
the
obj
ect
locat
ion
pro
ba
bili
ty
and
generate
d
t
he
loca
ti
on
pro
bab
il
it
y
m
ap
.
To
e
valuate
the
pe
rfo
rm
ance
of
t
he
pr
opos
e
d
m
et
ho
d,
we
ge
nera
te
d
the
gr
ound
trut
hs
of
t
he
98
ultraso
nogram
s
of
gallbla
dd
er
a
nd
a
ppli
ed
the
pro
pose
d
m
et
ho
d
t
o
gallbla
dder
de
te
ct
ion
.
E
xper
i
m
ental
resu
lt
s
sho
wed
98.
02%
s
ucce
ss
rate
by
ge
ne
rati
ng
the
ca
ndidate
s
inclu
di
ng
99
of
101
gallbla
dder
re
gions.
The pr
opose
d
m
et
ho
d
e
ff
e
ct
ively
locate
d ga
ll
bladd
e
r
in
u
lt
rason
ogram
an
d
ca
n be a
pp
li
e
d
in
v
a
rio
us
fie
lds
.
REFERE
NCE
S
[1]
S.
J.
Sree
an
d
C.
Vasan
thana
y
aki,
“
Ultra
sound
Fet
al
Im
age
Segm
ent
a
tion
T
ec
hniqu
es:
A
R
eview,
”
Curr
ent
Me
di
cal Imaging
R
evie
w
s
,
vol. 15, no. 1 pp.
52
-
60
,
2019
.
[2]
N.
Shrivasta
v
a
and
J.
Bharti,
“
A
Com
par
at
ive
Anal
y
sis
of
Medical
Im
ag
e
S
egmenta
t
ion,”
I
n
Inte
rnation
a
l
Confe
renc
e
on
A
dvanc
ed
Comput
ing
Ne
tworki
ng
and
Informatic
s
,
vol. 870, pp. 45
9
-
467,
2019
.
[3]
J.
A.
Hid
es,
et
a
l.
,
“
Us
e
of
r
ea
l
-
t
ime
ult
r
asound
i
m
agi
ng
f
or
f
ee
d
bac
k
in
r
eha
b
il
i
t
at
ion
,
”
Manual
t
herapy
,
vol.
3,
no.
3
,
pp
.
125
-
1
31,
1998
.
[4]
H.
J.
Park
,
e
t
al
.
,
“
An
eff
e
ct
iv
e
col
or
qu
ant
i
za
t
i
on
m
et
hod
usin
g
col
or
importa
nce
-
base
d
self
-
o
rga
nizing
m
aps,
”
Neural
Ne
twork Wor
ld
,
vol
.
25
,
n
o.
2
,
pp
.
121
-
13
7,
2015
.
[
5]
T.
Yam
ad
a,
et al
.
,
Text
boo
k
o
f
ga
stroente
rology
(
Vol
.
5)
.
W
il
e
y
-
B
la
ckwe
ll
,
2009.
[6]
T.
C.
Noone
,
e
t
al.
,
“
Abdom
ina
l
imaging
stud
ie
s:
compar
ison
of
dia
gnostic
accur
ac
i
es
r
esult
ing
fr
om
ult
rasoun
d
,
computed
tomo
gra
ph
y
,
and
m
a
gnet
i
c
reson
anc
e
imaging
in
t
h
e
sam
e
indi
vidu
al,
”
Magne
ti
c
resonance
imaging
,
vol.
22,
no.
1,
p
p.
19
-
24
,
2004
.
[7]
D.
Abraha
m
,
et
al.
,
“
Eme
rgen
c
y
Me
d
ic
in
e
Sono
graphy:
Pocke
t
Guide
to
Sonographic
Anat
omy
and
Pat
holog
y
,”
Jones &
Bar
tlett
Le
arn
ing, 2009.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
9
, N
o.
6
,
Dece
m
ber
2
01
9
:
5227
-
5234
5234
[8]
S.
Bodzi
och
and
M.
R.
Ogi
el
a
,
“
New
appr
oac
h
t
o
gal
lb
la
dde
r
ultrasonic
imag
es
a
naly
s
is
and
le
sio
ns
rec
ognition,
”
Computerized
M
edi
ca
l
Imaging
a
nd
Gr
aphic
s
,
vo
l
.
33
,
no
.
2
,
pp
.
1
54
-
170,
2009
.
[9]
S.
Bodz
ioc
h
,
“
Autom
at
ed
D
et
e
ct
ing
S
y
m
ptoms
of
Sel
ec
t
ed
Ga
ll
bla
dd
er
Il
lne
ss
Based
on
A
St
a
tic
Ul
tra
soun
d
Im
age
s Anal
y
s
is,”
Bio
-
Al
gorithm
s and
Me
d
-
S
ystem
s
,
vol. 2, no. 3, pp. 33
-
42,
2006
.
[10]
C.
T
.
Bo
ll
ig
er
,
e
t
al.
,
“
Cli
n
ical
Ch
est
Ultr
asound,
”
Progress
in
Re
s
piratory
R
ese
arc
h
,
vol
.
37
,
pp
.
18
2
-
188,
2009
.
[11]
V.
Munee
sw
ar
a
n
and
M.
P.
R
a
ja
seka
r
an,
“
Autom
at
ic
Segm
entati
on
of
Gall
b
ladder
Us
ing
Intu
it
ioni
st
ic
Fuz
z
y
Based
Act
ive Co
ntour
Model
,
”
M
ic
roel
ec
troni
cs,
El
e
ct
rom
agnet
i
c
s and
Telecomm
unic
ati
ons
,
pp
.
6
51
-
658,
2019
.
[12]
M.
Cie
cho
le
ws
ki,
“G
all
bladd
er
segment
ati
o
n
in
2
-
D
ultr
asound
images
using
def
orm
able
cont
our
methods,
”
Intern
at
ion
al
Confer
e
nce
on
Modeling
Dec
isions
f
or
Artif
ic
i
al
In
te
lligen
ce,
Spri
nger
,
Ber
li
n
,
Heide
lb
erg
,
201
0.
[13]
J.
Li
an
,
e
t
al.
,
“
Autom
at
ic
ga
ll
b
la
dder
and
ga
ll
s
tone
reg
ions
segm
ent
at
ion
i
n
ultrasound
image,”
Inte
rnat
iona
l
journal
of
comp
ute
r ass
iste
d
rad
iol
ogy
and
sur
ge
ry
,
vol
.
12
,
no
.
4
,
pp
.
553
-
568
,
2
017.
[14]
V.
Munee
sw
ar
a
n
and
M.
P
.
R
ajaseka
ran
,
“
Autom
at
ic
s
egmenta
t
ion
of
gallbla
dd
er
using
bio
-
ins
pire
d
al
gor
it
hm
base
d
on
a
spid
e
r
web c
onstructi
on
m
odel
,
”
The
Journal
of
Super
computi
ng
,
pp.
1
-
26,
2018
.
[15]
C.
Tomasi
and
R.
Mandu
chi,
“
Bi
lateral
filte
rin
g
for
gray
and
col
or
images
,
”
Proce
edi
ngs
of
the
1998
IE
E
E
Inte
rna
ti
ona
l
Co
nfe
ren
c
e
on
Co
m
pute
r
Vision,
p
p.
839
-
846
,
Jan
.
1998.
[16]
A.
Yoza
,
et
al
.
,
“
A
Stud
y
on
Eff
ec
t
ive
Repetitio
n
of
B
ilate
r
al
Fil
te
r
for
Med
ical
I
m
age
s,”
Bul
l
etin
of
N
et
working
,
Computing,
S
yst
ems,
and
So
ft
wa
re
,
vol
.
8
,
no
.
1
,
pp.
41
-
44
,
2019
.
[17]
T.
Huang
,
et
al
.
,
“
A
fast
two
-
di
m
ensiona
l
m
edian
filter
ing
a
lgo
rit
hm
,
”
IE
E
E
Tr
ansacti
ons
on
A
cousti
cs,
Sp
eech
,
and
Signal P
roc
essing
,
vol
.
27
,
n
o.
1
,
pp
.
13
-
18
,
1979.
[18]
R.
C.
Gonza
lez a
nd
R.
E. Woods
,
“
Digit
al image pr
oce
ss
ing
3rd e
dit
ion
,
”
Pren
tice Hall
,
Nueva
Jers
e
y
,
2008
.
[19]
M.
E
.
C
el
eb
i,
et
al.
,
“
An
eff
ec
t
iv
e
re
al
-
ti
m
e
col
or
quantiz
at
ion
m
e
thod
base
d
on
di
visive
h
ie
r
arc
hi
c
al
cl
uste
ring,”
Journal
of
R
eal
-
Time
Image Proce
ss
ing
,
vo
l. 10,
no.
2
,
pp
.
329
-
3
44,
2012
.
[20]
Y.
C.
Hu
and
B
.
H.
Su
,
“
Acc
e
le
r
at
ed
K
-
m
ea
ns
c
l
usteri
ng
al
gor
it
h
m
for
col
our
im
age
qu
antization
,
”
The
Imagi
ng
Sci
en
ce J
ournal
,
vol. 56, no. 1, p
p.
29
-
40
,
2008
.
[21]
G.
Schae
f
er
an
d
H.
Zhou,
“
Fuzz
y
c
luste
ri
ng
for
col
our
r
ed
uct
ion
in
image
s,”
Telecomm
unic
ati
on
S
yste
ms
,
vol.
40
,
no
.
1
-
2
,
pp.
17
-
25
,
2009
.
[22]
Q.
W
en
and
M.
E.
Celebi
,
“
Hard
ver
sus
Fuz
z
y
c
-
m
ea
ns
cl
ust
eri
ng
for
col
or
qu
ant
i
za
t
ion,
”
EURA
S
IP
Jo
urnal
o
n
Adv
anc
es
in
Sig
nal
Proc
essing
,
vol.
2011
,
no
.
1
,
pp.
118
-
129
,
20
11.
[23]
K.
L.
Chung,
et
al
.
,
“
Speedu
p
of
co
lor
pa
l
et
t
e
ind
exi
ng
i
n
self
-
orga
n
izat
ion
of
kohon
en
feature
m
ap,
”
Ex
pert
Syste
ms
wit
h
App
licati
on
s
,
vol. 39, no. 3,
pp.
2427
-
2432
,
2012.
[24]
J.
Rast
i,
e
t
a
l.
,
“
Color
red
u
ct
io
n
using
a
m
ult
i
-
stage
kohonen
self
-
orga
ni
zi
ng
m
ap
with
red
un
dant
fe
at
ure
s
,
”
Ex
pert
Syste
ms
wit
h
App
licati
on
s
,
vol. 38, n
o.
10
,
pp
.
13188
-
1319
7,
2011
.
[25]
G.
Schae
f
er,
“
Inte
lligent
approache
s
to
col
our
pale
t
te
d
esign
,
”
Innova
ti
ons
i
n
Inte
l
li
g
ent
Im
age
Ana
l
y
s
is,
Springer
Ber
li
n
Heide
lb
erg
,
pp.
275
-
289,
2011
.
[26]
T.
Yam
ad
a
,
et al
.
,
“
Text
book
of
g
astroente
rology
,”
(vol.
5
), W
i
ley
-
Bla
ckwe
ll
,
200
9.
[27]
D.
Abraha
m
,
e
t
al.
,
“
Emerge
nc
y
Medicine
Sonograph
y
:
Pock
et
Guide
to
Sonogr
aphi
c
Anatom
y
and
Patho
lo
g
y
,
”
Jone
s
&
Bartl
et
t
Learning
,
2009
.
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Hy
un
Jun
Park
,
He
recei
ved
his
M.S.
and
Ph.D.
degr
ee
s
fro
m
the
Depa
rtme
nt
of
Com
pute
r
Engi
ne
eri
ng,
Pus
an
Nati
ona
l
Univer
sit
y
,
Busan,
Korea
,
in
2009
a
nd
2017,
r
espe
ct
iv
ely
.
In
2017,
he
was
a
postdo
ct
ora
l
rese
arc
h
er
at
BK21P
LUS,
Crea
t
ive
Hum
an
Resourc
e
Dev
elopm
ent
Program
for
IT
Converg
enc
e
,
Pus
an
Na
ti
onal
Univer
sit
y
,
Korea
.
From
2018
to
th
e
p
rese
nt,
he
is
an
associa
t
e
p
rofe
s
sor
at
the
Divis
ion
of
Softwar
e
Converg
ence,
Cheongj
u
Unive
rsit
y
,
Korea
.
Hi
s
rese
arc
h
in
te
rest
s
inc
lud
e
computer
v
ision,
image
proc
essing,
fa
ctor
y
aut
o
m
ation,
neur
al
net
work
,
and
de
ep le
a
rnin
g
applications.
K
w
ang
Ba
ek
K
im
,
Kw
ang
Ba
e
k
Kim
recei
ved
his
M.S.
and
Ph.
D.
degr
ee
s
from
the
Depa
r
tment
of
Com
pute
r
Sci
enc
e
,
Pus
an
Nat
iona
l
Univer
si
t
y,
Busan
,
Kor
ea,
in
1993
and
199
9,
r
espe
ctive
l
y
.
From
1997
to
the
pr
ese
nt
,
h
e
is
a
prof
essor
at
th
e
Div
isio
n
of
Com
pute
r
and
Inform
at
io
n
Engi
ne
eri
ng,
Sill
a
Univer
sit
y
,
Ko
rea
.
He
is
cur
r
en
tly
an
associate
edi
tor
for
Journ
a
l
of
Intelli
g
ence
and
Inform
ation
S
y
st
ems
a
nd
T
he
Open
Com
pute
r
Scie
n
ce
Jour
nal
(US
A).
His
rese
arc
h
int
er
est
s
inc
lud
e
fuz
z
y
n
e
ura
l
n
et
work
and
applications, bi
oinformati
cs,
an
d
image
proc
essi
ng.
Evaluation Warning : The document was created with Spire.PDF for Python.