T
E
L
KO
M
N
I
KA
T
e
lec
om
m
u
n
icat
ion
,
Com
p
u
t
i
n
g,
E
lec
t
r
on
ics
an
d
Cont
r
ol
Vol.
18
,
No.
4
,
Augus
t
2020
,
pp.
2213~2223
I
S
S
N:
1693
-
6930,
a
c
c
r
e
dit
e
d
F
ir
s
t
G
r
a
de
by
Ke
me
nr
is
tekdikti
,
De
c
r
e
e
No:
21/E
/KP
T
/2018
DO
I
:
10.
12928/
T
E
L
KO
M
NI
KA
.
v18i4.
14228
2213
Jou
r
n
al
h
omepage
:
ht
tp:
//
jour
nal.
uad
.
ac
.
id/
index
.
php/T
E
L
K
OM
N
I
K
A
D
e
t
e
c
t
io
n
of
u
ve
al
m
e
la
n
om
a
u
si
n
g f
u
z
z
y an
d
n
e
u
r
al
n
e
t
w
or
k
s c
la
ssi
f
ie
r
s
Dani
e
l
F
.
S
an
t
os
,
Hel
b
e
r
t
E
.
E
s
p
it
ia
Facu
l
t
ad
d
e
I
n
g
e
n
i
er
í
a
,
U
n
i
v
er
s
i
d
ad
D
i
s
t
ri
t
a
l
Fran
ci
s
co
J
o
s
é
d
e
Ca
l
d
a
s
,
Bo
g
o
t
á,
Co
l
o
m
b
i
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
S
e
p
28,
2019
R
e
vis
e
d
M
a
r
4,
2020
Ac
c
e
pted
M
a
r
27,
2020
T
h
e
u
s
e
o
f
i
mag
e
p
r
o
ces
s
i
n
g
i
s
i
n
crea
s
i
n
g
l
y
u
t
i
l
i
ze
d
fo
r
d
i
s
ea
s
e
d
et
ec
t
i
o
n
.
In
t
h
i
s
art
i
cl
e,
an
al
g
o
r
i
t
h
m
i
s
p
r
o
p
o
s
ed
t
o
d
et
ec
t
u
v
e
al
m
el
an
o
ma
(U
M)
w
h
i
ch
i
s
a
t
y
p
e
o
f
i
n
t
ra
o
cu
l
ar
can
cer.
T
h
e
p
ro
p
o
s
ed
me
t
h
o
d
i
n
t
eg
ra
t
es
al
g
o
r
i
t
h
ms
rel
at
e
d
t
o
i
r
i
s
s
e
g
men
t
at
i
o
n
an
d
p
ro
p
o
s
es
a
n
o
v
e
l
al
g
o
r
i
t
h
m
f
o
r
t
h
e
d
et
ec
t
i
o
n
o
f
U
M
fr
o
m
t
h
e
ap
p
ro
ac
h
o
f
f
u
zzy
l
o
g
i
c
a
n
d
n
e
u
ral
n
et
w
o
r
k
s
.
T
h
e
s
t
u
d
y
ca
s
e
res
u
l
t
s
s
h
o
w
76%
co
rrect
cl
as
s
i
fi
ca
t
i
o
n
i
n
t
h
e
fu
zzy
l
o
g
i
c
s
y
s
t
em
an
d
96
.
04%
fo
r
t
h
e
art
i
fi
c
i
al
n
eu
ra
l
n
et
w
o
rk
s
.
K
e
y
w
o
r
d
s
:
F
uz
z
y
S
ys
tems
I
mage
pr
oc
e
s
s
ing
Ne
ur
a
l
ne
twor
ks
Uve
a
l
mela
noma
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e
.
C
or
r
e
s
pon
din
g
A
u
th
or
:
He
lber
t
E
.
E
s
pit
ia
,
F
a
c
ult
a
d
de
I
nge
nier
ía
,
Unive
r
s
idad
Dis
tr
it
a
l
F
r
a
nc
is
c
o
J
os
é
de
C
a
ldas
,
B
ogotá
,
C
olom
bia
.
E
mail:
he
e
s
pit
iac
@udis
tr
it
a
l.
e
du.
c
o
1.
I
NT
RODU
C
T
I
ON
T
he
tr
a
dit
ional
im
a
ge
pr
oc
e
s
s
ing
is
us
e
d
to
s
olve
pr
oblems
li
ke
qua
li
ty
im
pr
ove
ment
,
r
e
s
tor
a
ti
on
,
highl
ight
f
e
a
tur
e
s
,
a
mong
other
s
.
C
ur
r
e
ntl
y
,
with
the
e
mer
ge
nc
e
of
br
a
nc
he
s
s
uc
h
a
s
f
uz
z
y
logi
c
a
n
d
ne
ur
a
l
ne
twor
ks
,
im
a
ge
pr
oc
e
s
s
ing
is
a
ls
o
be
ing
us
e
d
to
a
ddr
e
s
s
pr
oblems
r
e
late
d
to
dis
e
a
s
e
s
.
Among
them
is
the
uve
a
l
mela
noma
(
UM
)
,
a
s
ubtype
of
oc
ular
mela
noma
(
OM
)
[
1
]
a
type
of
int
r
a
oc
ular
c
a
nc
e
r
that
a
r
is
e
s
in
the
mela
noc
ytes
of
the
c
olour
e
d
pa
r
t
o
f
the
e
ye
(
i
r
is
)
,
c
ir
c
le
of
mus
c
le
ti
s
s
ue
(
c
il
iar
y
body)
o
r
in
th
e
lar
ge
s
t
pa
r
t
of
the
uve
a
t
r
a
c
k
(
c
hor
oid
)
whic
h
is
loca
ted
b
e
ne
a
th
the
r
e
ti
na
in
the
ba
c
k
pa
r
t
of
the
e
ye
.
M
os
tl
y,
oc
ular
mela
no
mas
a
ppe
a
r
withi
n
the
c
ho
r
oid
a
nd
c
ome
f
r
om
the
mela
noc
ytes
,
whic
h
a
r
e
the
c
e
ll
s
of
the
b
ody
that
pr
oduc
e
pigm
e
nt
[
2]
.
E
a
r
ly
diagnos
is
a
nd
loca
l
tr
e
a
tm
e
nt
a
r
e
c
r
uc
ial
s
ince
s
ur
vival
r
a
tes
c
or
r
e
late
with
the
s
ize
of
the
pr
im
a
r
y
tu
mou
r
.
How
e
ve
r
,
a
ppr
o
xim
a
tely
50%
of
pa
ti
e
nts
will
de
ve
lop
meta
s
tas
e
s
wi
th
a
s
ur
vival
r
a
te
of
6
-
12
mont
hs
f
r
om
diagnos
is
[
3
]
.
Us
ing
moder
n
tool
s
li
ke
ult
r
a
s
onogr
a
phy,
f
l
uor
e
s
c
e
in
a
ngiogr
a
phy,
a
nd
opti
c
a
l
c
ohe
r
e
nc
e
tom
ogr
a
phy
c
a
n
s
igni
f
ica
ntl
y
a
id
in
diagnos
is
[
4
,
5
]
,
r
oughly
3
out
o
f
4
pe
ople
with
oc
ular
mela
noma
s
ur
vive
f
or
a
t
lea
s
t
5
ye
a
r
s
.
S
ur
vival
r
a
tes
tend
to
be
be
tt
e
r
f
or
c
a
nc
e
r
de
tec
ti
on
in
e
a
r
li
e
r
s
tage
s
than
f
or
thos
e
in
late
r
s
tage
s
.
W
he
n
c
a
nc
e
r
is
c
onf
ined
to
the
e
ye
,
the
r
e
lative
s
ur
vival
r
a
te
a
t
5
ye
a
r
s
is
a
r
ound
80%
.
F
or
pe
ople
with
e
ye
mela
nomas
that
ha
ve
s
pr
e
a
d
to
dis
tant
p
a
r
ts
of
the
body
,
the
5
-
ye
a
r
r
e
lative
s
ur
vival
r
a
te
i
s
a
r
ound
15%
[
6]
a
nd
1
-
ye
a
r
s
ur
vival
r
a
te
is
10%
to
15%
[7
–
9]
.
I
n
r
e
late
d
wor
ks
,
a
lar
ge
number
of
p
r
ojec
ts
a
s
s
oc
iate
d
with
the
i
r
is
ha
ve
be
e
n
de
ve
loped.
T
he
r
e
a
r
e
two
main
a
ppli
c
a
ti
ons
:
identif
ica
ti
on
o
f
pe
ople
t
hr
ough
unique
pa
tt
e
r
ns
[
10
–
12
]
a
nd
de
tec
ti
on
of
dis
e
a
s
e
s
whe
r
e
I
r
idol
ogy
is
f
ound
,
whic
h
is
a
br
a
nc
h
o
f
a
lt
e
r
na
ti
ve
medic
ine
that
is
r
e
s
pons
ibl
e
f
or
e
xa
mi
ning
pa
t
ter
ns
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
4
,
Augus
t
2020
:
2213
-
2223
2214
c
olour
s
,
a
nd
other
c
ha
r
a
c
ter
is
ti
c
s
of
the
ir
is
to
de
ter
mi
ne
the
he
a
lt
h
of
pa
ti
e
nts
[
13,
14]
.
F
o
r
e
xa
m
ple
[
15]
,
whe
r
e
the
ir
is
is
e
mpl
oye
d
f
o
r
the
de
tec
ti
on
of
Alz
he
im
e
r
.
I
n
[
16]
is
s
hown
how
to
de
tec
t
dis
e
a
s
e
s
in
the
ir
is
us
ing
the
Ga
bor
f
il
te
r
.
Am
ong
the
c
las
s
if
ied
dis
e
a
s
e
s
a
r
e
c
or
ne
a
l
oe
de
ma,
ir
idot
omi
e
s
,
a
nd
c
onjunctivi
ti
s
.
M
or
e
ove
r
,
in
[
17]
,
thr
ough
tec
hniques
of
I
r
idol
ogy
a
nd
the
W
a
ter
s
he
d
tr
a
ns
f
or
m
dis
e
a
s
e
s
a
r
e
identif
ied,
r
e
a
c
hing
a
n
a
c
c
ur
a
c
y
of
87
.
5%
in
the
de
tec
ti
on
of
kidney
p
r
oblems
.
I
n
[
18]
,
us
ing
un
ique
ir
is
f
e
a
tur
e
s
a
nd
a
dva
nc
e
d
e
nc
r
ypti
on
s
tand
a
r
d
(
AE
S
)
im
a
ge
s
a
r
e
e
nc
r
ypted,
a
c
hieving
high
leve
ls
of
s
e
c
ur
it
y.
As
s
hown
in
[
2
,
5]
ther
e
is
no
a
ppr
ove
d
ther
a
py
to
tr
e
a
t
meta
s
tatic
uve
a
l
mela
noma
a
nd
c
ons
ider
ing
that
it
s
e
a
r
ly
de
tec
ti
on
s
igni
f
ica
ntl
y
incr
e
a
s
e
s
th
e
c
ha
nc
e
s
of
s
ur
vival,
ther
e
wa
s
de
v
e
loped
a
m
e
thod
to
identif
y
UM
.
T
he
p
r
opos
e
d
method
int
e
gr
a
tes
a
lgor
it
hms
us
e
d
in
wor
k
r
e
late
d
to
ir
is
s
e
gmenta
ti
on
a
nd
pr
opos
e
s
a
methodology
f
or
UM
de
tec
ti
on
f
r
om
the
f
uz
z
y
logi
c
a
nd
ne
ur
a
l
ne
twor
k
a
ppr
oa
c
h,
a
s
we
ll
a
s
pr
e
s
e
nt
a
va
r
iation
of
the
a
lgor
it
hm
pr
opos
e
d
by
W
il
de
s
in
[
19]
.
T
he
va
r
iation
im
pleme
nts
a
ddit
ion
a
l
logi
c
to
incr
e
a
s
e
the
a
c
c
ur
a
c
y
a
s
we
ll
a
s
im
pleme
nti
ng
da
ta
s
tr
uc
tur
e
s
li
ke
a
r
e
d
-
blac
k
tr
e
e
a
nd
dis
joi
nt
s
e
t
with
the
union
by
r
a
nk
a
nd
pa
th
c
ompr
e
s
s
ion
to
r
e
duc
e
t
he
tur
na
r
ound
ti
me.
T
he
s
hown
a
lgo
r
it
hm
e
mpl
oye
d
to
de
tec
t
UM
c
ons
is
ts
of
:
1)
p
r
e
pr
oc
e
s
s
im
a
ge
s
with
a
nd
wi
thout
UM
,
2)
a
pply
a
s
e
gmenta
ti
on
a
lgor
i
thm
to
e
a
c
h
o
f
thes
e
pr
e
p
r
oc
e
s
s
e
d
im
a
ge
s
to
f
ind
the
r
e
gion
of
int
e
r
e
s
t
(
R
OI
)
,
3)
tr
a
ns
f
or
m
the
R
OI
to
a
nother
s
pa
c
e
that
a
ll
ows
a
n
a
na
lys
is
with
les
s
nois
e
,
4)
obtain
de
s
c
r
ipt
or
s
of
the
R
OI
s
to
ge
ne
r
a
te
a
tr
a
ini
ng
da
tas
e
t,
5)
a
f
u
z
z
y
a
nd
a
ne
ur
a
l
ne
twor
k
c
las
s
if
ier
a
r
e
c
r
e
a
ted
ba
s
e
d
on
the
tr
a
ini
ng
da
tas
e
t,
a
nd
f
inally
in
6)
the
r
e
s
ult
s
of
t
he
c
las
s
if
ier
s
a
r
e
a
na
lys
e
d
a
nd
c
ompar
e
d.
2.
I
M
AGE
P
ROCE
S
S
I
NG
Huma
ns
a
r
e
ve
r
y
good
a
t
de
tec
ti
ng
pa
tt
e
r
ns
but
unli
ke
mac
hines
,
we
a
r
e
not
good
a
t
pr
oc
e
s
s
ing
lar
ge
a
mount
s
o
f
da
ta,
then,
i
f
the
mac
hine
is
ta
ught
to
identi
f
y
pa
tt
e
r
ns
it
is
pos
s
ibl
e
to
a
utom
a
te
many
s
pe
c
ialize
d
tas
ks
s
uc
h
a
s
dis
e
a
s
e
de
tec
ti
on.
F
or
t
he
pr
e
s
e
nt
c
a
s
e
s
tudy,
Hough
c
i
r
c
ular
tr
a
ns
f
or
m
,
a
da
pti
ve
binar
iza
ti
ons
,
f
il
ter
s
,
Hu
mom
e
nts
,
a
mong
other
s
t
e
c
hniques
a
r
e
us
e
d
to
tr
a
in
the
c
las
s
if
ier
s
to
identif
y
UM
,
a
br
ief
de
s
c
r
ipt
ion
o
f
thos
e
a
lgo
r
it
hms
is
pr
ovided
.
2.
1.
Hou
gh
c
irc
u
lar
t
r
an
s
f
or
m
Hough
c
ir
c
ular
t
r
a
ns
f
or
m
(
HC
T
)
is
a
n
a
lgor
it
h
m
to
s
e
a
r
c
h
c
i
r
c
les
in
im
a
ge
s
,
thi
s
a
ppr
oa
c
h
is
us
e
d
be
c
a
us
e
it
pr
ovides
r
obus
tnes
s
to
the
pr
e
s
e
nc
e
of
nois
e
,
oc
c
lus
ion
a
nd
va
r
iation
of
il
lum
ination
[
2
0]
.
T
his
a
lgor
it
hm
ha
s
thr
e
e
f
unda
menta
l
pha
s
e
s
:
1.
Ac
c
umul
a
tor
a
r
r
a
y
c
omput
a
ti
on
2.
C
e
nter
e
s
ti
matio
n
3.
R
a
dius
e
s
ti
mation
T
he
HC
T
is
us
e
d
to
tr
a
ns
f
or
m
a
s
e
t
of
c
ha
r
a
c
ter
is
ti
c
s
of
point
s
in
the
s
pa
c
e
of
the
im
a
ge
,
(
1)
to
a
s
e
t
of
votes
a
c
c
umul
a
ted
in
the
s
pa
c
e
of
the
pa
r
a
mete
r
s
(
2)
.
T
he
s
e
votes
a
r
e
int
e
ge
r
va
lues
that
a
r
e
s
tor
e
d
in
a
n
a
r
r
a
y.
T
he
pos
it
ion
of
the
a
r
r
a
nge
ment
tha
t
c
ontai
ns
the
mos
t
votes
indi
c
a
tes
the
pr
e
s
e
nc
e
of
a
c
ir
c
le.
T
he
(
1)
r
e
pr
e
s
e
nts
a
c
ir
c
le,
whe
r
e
is
the
r
a
dius
a
nd
(
0
,
0
)
a
r
e
the
c
oor
dinate
s
of
the
c
e
ntr
e
o
f
the
c
i
r
c
le.
2
=
(
−
0
)
2
+
(
−
0
)
2
(
1)
T
he
(
2
)
r
e
pr
e
s
e
nts
the
pa
r
a
metr
iza
ti
on
o
f
the
c
i
r
c
le
whe
r
e
(
0
,
0
)
a
r
e
the
c
oor
dinate
s
o
f
the
c
e
ntr
e
o
f
the
c
ir
c
le
a
nd
is
the
a
ngle
of
inclinat
ion.
T
ha
t
e
qu
a
ti
on
a
ll
ows
tr
a
ns
f
or
mi
ng
a
c
ir
c
le
int
o
a
r
e
c
tangle
.
=
0
+
co
s
(
)
=
0
+
s
in
(
)
(
2)
2
.
2.
A
d
ap
t
ive
t
h
r
e
s
h
old
in
g
C
onve
nti
ona
l
thr
e
s
holdi
ng
methods
us
e
a
th
r
e
s
hold
f
or
a
ll
pixels
,
while
a
da
pti
ve
thr
e
s
holdi
ng
(
A
T
)
c
ha
nge
s
the
thr
e
s
hold
va
lue
dyna
mi
c
a
ll
y
on
the
im
a
ge
.
T
he
AT
ha
s
s
hown
be
tt
e
r
r
e
s
ult
s
with
r
e
s
pe
c
t
to
the
tr
a
dit
ional
th
r
e
s
holdi
ng
s
ince
the
il
lum
inatio
n
a
nd
the
s
ha
dows
c
ha
nge
de
pe
nding
on
the
pos
it
ion
of
the
im
a
ge
[
21
]
.
W
it
hin
the
thr
e
s
holdi
ng
methods
us
e
d
in
thi
s
a
r
ti
c
le
a
r
e
:
Ga
us
s
ian
thr
e
s
holdi
ng
(
r
e
pr
e
s
e
nted
by
(
3)
)
a
nd
mea
n
thr
e
s
holdi
ng.
I
n
(
4)
is
de
picte
d
a
n
e
xa
mpl
e
of
a
Ga
us
s
ian
window.
ℎ
(
,
)
=
1
2
2
−
2
+
2
2
2
(
3)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
tec
ti
on
of
uv
e
al
me
lanoma
us
ing
fuz
z
y
and
ne
ur
al
ne
tw
or
k
s
c
las
s
if
ie
r
s
(
Danie
l
F
.
Santos
)
2215
=
1
16
[
1
2
1
2
4
2
1
2
1
]
(
4)
2.
3.
Hu
m
om
e
n
t
s
T
he
Hu
M
oments
a
ll
ow
to
obtain
im
a
ge
f
e
a
tur
e
s
us
e
d
in
the
c
las
s
if
ica
ti
on
pr
oc
e
s
s
.
T
he
f
oll
owing
de
f
ini
ti
on
is
im
po
r
tant
in
thi
s
r
e
ga
r
d.
De
f
ini
ti
on
1
(
M
ome
nt
of
an
image
)
:
A
me
as
ur
e
that
pr
ov
ides
a
ge
ne
r
ic
r
e
p
r
e
s
e
ntat
ion
of
an
obj
e
c
t,
w
it
h
s
imple
or
c
omple
x
fi
gur
e
s
[
21]
.
T
he
r
e
a
r
e
a
c
ons
ider
a
ble
number
o
f
us
e
d
a
nd
we
ll
-
known
mom
e
nts
withi
n
whic
h
we
c
a
n
include
ge
ometr
ic
mom
e
nts
[
23]
,
Z
e
r
nike
mom
e
nts
[
24]
,
r
otational
mom
e
nts
[
25]
,
a
nd
c
ompl
e
x
mom
e
n
ts
[
26]
.
T
he
invar
iants
o
f
the
mom
e
nt
we
r
e
int
r
oduc
e
d
b
y
Hu
in
[
23]
,
thes
e
a
r
e
ve
r
y
us
e
f
ul
pr
ope
r
ti
e
s
tha
t
c
a
n
be
e
xtr
a
c
ted
f
r
om
a
n
im
a
ge
be
c
a
us
e
they
a
r
e
not
onl
y
indepe
nde
nt
to
pos
it
ion,
s
ize
a
nd
or
ienta
ti
on
bu
t
a
ls
o
to
p
a
r
a
ll
e
l
pr
ojec
ti
on,
thus
they
ha
ve
be
e
n
wide
ly
u
s
e
d
to
pe
r
f
or
m
pa
tt
e
r
n
r
e
c
ognit
ion,
im
a
ge
r
e
gis
tr
a
ti
on
a
nd
im
a
ge
r
e
c
ons
tr
uc
ti
on
[
23
,
27
–
29
]
.
2.
4.
I
n
var
ian
t
m
om
e
n
t
s
A
two
-
dim
e
ns
ional
mom
e
nt
of
o
r
de
r
(
+
)
is
de
f
ined
by
the
(
5
)
.
=
∫
∞
−
∞
∫
∞
−
∞
(
)
(
)
(
,
)
,
=
0
,
1
,
2
.
.
.
(
5)
I
f
the
f
unc
ti
on
(
,
)
is
de
f
ined
in
pa
r
ts
,
the
mom
e
nts
of
a
ll
o
r
de
r
s
e
xis
t
a
nd
the
s
e
que
nc
e
of
mom
e
nts
is
only
de
ter
mi
ne
d
by
the
f
unc
ti
on
(
,
)
,
but
thos
e
mo
ments
in
the
(
5)
may
not
be
invar
iant
to
tr
a
ns
latio
n,
r
otation
or
s
c
a
le,
thus
the
invar
iant
Hu
mom
e
nts
c
a
n
be
c
a
lcula
ted
us
ing
the
c
e
ntr
a
l
mom
e
nts
,
whic
h
a
r
e
de
f
ined
by
(
6
)
.
=
∫
∞
−
∞
∫
∞
−
∞
(
−
′
)
(
−
′
)
(
,
)
,
=
0
,
1
,
2
…
(
6
)
T
he
mom
e
nts
a
r
e
c
a
lcula
ted
u
s
ing
the
i
mage
(
,
)
,
the
s
e
a
r
e
e
quivale
nt
to
whos
e
c
e
ntr
e
ha
s
be
e
n
moved
to
the
c
e
ntr
oid
of
the
im
a
ge
,
ther
e
f
or
e
,
the
mom
e
nts
a
r
e
invar
iant
to
tr
a
ns
lation
[
23]
.
W
he
r
e
t
he
pixel
(
′
,
′
)
is
the
c
e
ntr
oid
of
the
im
a
ge
a
nd:
′
=
10
00
,
′
=
01
00
W
it
h
the
us
e
of
the
no
r
maliza
ti
on
o
f
the
mom
e
nts
the
invar
ianc
e
to
s
c
a
le
c
a
n
be
obtaine
d.
T
his
mo
ment
c
a
n
be
de
f
ined
by:
=
00
,
=
+
+
2
2
,
+
=
2
,
3
,
…
(
7)
B
a
s
e
d
on
thos
e
mom
e
nts
,
Hu
in
[
23]
pr
e
s
e
nted
the
s
e
ve
n
mom
e
nt
invar
iants
s
hown
be
low:
ℎ
1
=
20
+
02
ℎ
2
=
(
20
−
02
)
2
+
4
11
2
ℎ
3
=
(
30
−
3
12
)
2
+
(
3
21
−
03
)
2
ℎ
4
=
(
30
+
12
)
2
+
(
21
−
03
)
2
ℎ
5
=
(
30
−
3
12
)
(
30
+
12
)
[
(
30
+
12
)
2
−
3
(
21
−
03
)
2
]
+
(
3
21
−
03
)
(
21
+
03
)
[
3
(
30
+
12
)
2
−
(
21
+
03
)
2
]
ℎ
6
=
(
20
−
02
)
[
(
30
+
12
)
2
−
(
21
+
03
)
2
]
+
4
11
(
30
+
12
)
(
21
+
03
)
ℎ
7
=
(
3
21
−
03
)
(
30
+
12
)
[
(
30
+
12
)
2
−
3
(
21
−
03
)
2
]
−
(
30
−
3
12
)
(
21
+
03
)
[
3
(
30
+
12
)
2
−
(
21
+
03
)
2
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
4
,
Augus
t
2020
:
2213
-
2223
2216
I
t
is
im
po
r
tant
to
mention
that
the
no
is
e
ge
ne
r
a
te
d
by
dif
f
e
r
e
nt
f
a
c
tor
s
,
s
uc
h
a
s
the
type
o
f
c
a
mer
a
a
nd
f
il
e
f
or
matti
ng,
c
a
n
pr
oduc
e
e
r
r
or
s
whe
n
c
a
lcula
ti
ng
mom
e
nts
.
I
n
other
wor
ds
,
mom
e
nts
c
a
n
v
a
r
y
with
the
ge
ometr
ic
tr
a
ns
f
or
mation
of
the
im
a
ge
[
29
]
.
I
n
the
c
a
s
e
of
a
s
c
a
le
tr
a
ns
f
or
mation
or
r
otation
of
th
e
im
a
ge
,
a
r
ounding
of
pixel
pos
it
ions
o
r
int
e
r
polation
is
ge
ne
r
a
ted
whic
h
c
a
us
e
s
that
a
t
a
digi
tal
leve
l
the
inva
r
iants
of
the
mom
e
nt
a
ls
o
c
ha
nge
[
30
]
.
3.
RE
S
E
AR
CH
M
E
T
HO
D
T
his
s
e
c
ti
on
de
s
c
r
ibes
the
a
lgor
it
hm
(
s
e
e
F
igur
e
1)
,
e
a
c
h
s
ub
-
s
e
c
ti
on
r
e
pr
e
s
e
nts
e
a
c
h
s
tage
a
s
f
oll
ows
:
the
f
ir
s
t
one
c
or
r
e
s
ponds
to
the
im
a
ge
pr
oc
e
s
s
ing;
ne
xt,
r
e
f
e
r
s
to
the
de
tec
ti
on
of
the
r
e
gion
of
int
e
r
e
s
t;
a
f
ter
wa
r
ds
c
omes
the
f
i
lt
r
a
ti
on
a
nd
t
r
a
ns
f
or
mation
of
thos
e
r
e
gions
to
f
inally
pe
r
f
or
m
the
f
e
a
tur
e
e
xtr
a
c
ti
on.
F
igur
e
1.
S
tage
s
f
o
r
im
a
ge
f
e
a
tur
e
e
xtr
a
c
ti
on
3.
1.
I
m
age
p
r
e
-
p
r
oc
e
s
s
in
g
T
his
is
the
f
ir
s
t
pha
s
e
a
nd
it
is
of
vit
a
l
im
por
tanc
e
to
f
a
c
il
it
a
te
the
de
tec
ti
on
of
the
R
OI
,
in
whic
h
e
nter
s
a
n
im
a
ge
(
,
)
whos
e
output
is
the
im
a
ge
(
,
)
.
F
igur
e
2
s
hows
the
s
tage
s
thr
ough
whic
h
the
im
a
ge
pa
s
s
e
s
.
Us
ing
F
igur
e
3
it
is
s
hown
how
thi
s
s
tage
wor
ks
,
thi
s
im
a
ge
c
or
r
e
s
ponds
to
a
n
e
ye
a
f
f
e
c
ted
with
uve
a
l
mela
noma
[
31
]
.
T
he
s
e
que
nc
e
of
a
lgor
it
hms
that
f
a
c
il
it
a
te
the
de
tec
ti
on
of
the
R
OI
:
-
Gr
a
ys
c
a
le
tr
a
ns
f
or
mation:
F
ir
s
t,
a
n
im
a
ge
with
th
r
e
e
c
ha
nne
ls
R
GB
is
r
e
c
e
ived
a
nd
tr
a
ns
f
or
med
int
o
a
gr
a
ys
c
a
le
im
a
ge
;
thi
s
is
mainly
due
to
the
f
a
c
t
th
a
t
a
g
r
a
ys
c
a
le
im
a
ge
is
e
a
s
ier
to
pr
oc
e
s
s
than
a
n
R
GB
im
a
ge
(
s
e
e
F
igur
e
4
)
.
-
Apply
media
n
f
il
ter
:
T
he
im
a
ge
is
s
moot
he
d
by
a
pplyi
ng
a
media
n
f
il
ter
,
thi
s
will
de
c
r
e
a
s
e
the
n
ois
e
,
he
lpi
ng
to
make
the
R
OI
c
lea
r
e
r
.
T
he
r
e
s
ult
o
f
thi
s
pr
oc
e
s
s
is
s
hown
in
F
igur
e
5
.
-
Apply
binar
iza
ti
on:
W
it
h
th
is
s
tage
the
bound
a
r
ies
a
r
e
c
lea
r
ly
de
f
ined,
the
binar
iza
ti
on
us
e
d
wa
s
the
Ada
pti
ve
M
e
a
n
T
hr
e
s
holdi
ng
a
nd
the
Ada
pti
ve
Ga
us
s
ian
T
hr
e
s
holdi
ng.
F
igur
e
6
s
hows
the
r
e
s
ult
obtain
e
d
in
thi
s
s
tep.
-
I
mage
dil
a
ti
on:
Aga
in,
a
pr
oc
e
s
s
is
done
to
r
e
duc
e
nois
e
,
in
thi
s
c
a
s
e
,
pe
ppe
r
type
nois
e
,
us
ing
dil
a
ti
on.
T
he
r
e
s
pe
c
ti
ve
r
e
s
ult
obtaine
d
in
thi
s
s
tep
is
s
hown
in
F
igu
r
e
7
.
F
igur
e
2.
T
he
s
e
que
nc
e
of
a
lgor
it
hms
that
f
a
c
il
it
a
te
the
de
te
c
ti
on
o
f
the
R
OI
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
tec
ti
on
of
uv
e
al
me
lanoma
us
ing
fuz
z
y
and
ne
ur
al
ne
tw
or
k
s
c
las
s
if
ie
r
s
(
Danie
l
F
.
Santos
)
2217
F
igur
e
3.
E
ye
a
f
f
e
c
ted
with
uve
a
l
mela
noma
[
31
]
F
igur
e
4.
I
mage
with
UM
c
onve
r
ted
to
gr
a
ys
c
a
le
F
igur
e
5.
I
mage
with
UM
in
gr
a
ys
c
a
le
s
moot
he
d
us
ing
a
media
n
f
il
te
r
F
igur
e
6.
I
mage
with
UM
a
f
ter
a
pplyi
ng
a
da
pti
ve
mea
n
thr
e
s
holdi
ng
F
igur
e
7.
Dilate
d
i
mage
3.
2.
D
e
t
e
c
t
ion
o
f
t
h
e
r
e
gion
o
f
i
n
t
e
r
e
s
t
T
he
r
e
gion
o
f
int
e
r
e
s
t
is
the
ir
is
,
thi
s
is
a
ls
o
a
c
r
uc
ial
s
tage
f
or
the
de
tec
ti
on
o
f
UM
s
ince
a
n
inappr
opr
iate
loca
ti
on
of
the
R
OI
c
a
n
r
e
s
ult
i
n
e
xtr
a
c
ti
on
of
c
ha
r
a
c
ter
is
ti
c
s
f
r
om
non
-
r
e
leva
nt
pa
r
ts
,
ge
ne
r
a
ti
ng
a
tr
a
ini
ng
with
nois
e
a
nd
ther
e
f
or
e
a
n
e
r
r
on
e
ous
c
las
s
if
ica
ti
on.
B
a
s
e
d
on
the
li
ter
a
tu
r
e
of
i
r
is
s
e
gmenta
ti
on
a
lgor
it
hms
,
it
wa
s
de
c
ided
to
us
e
a
va
r
iation
of
the
w
il
de
s
a
lgor
it
hm
due
to
it
s
e
a
s
e
of
im
pleme
ntation,
c
omput
a
ti
ona
l
e
f
f
icie
nc
y
a
nd
s
uit
a
ble
pr
ope
r
ti
e
s
.
-
Algor
it
hm
o
f
R
OI
de
tec
ti
on
Algor
it
h
m
1
r
e
c
e
ives
the
im
a
ge
o
f
F
igur
e
7,
c
a
lcul
a
tes
the
width
a
nd
he
ight
of
the
im
a
ge
to
make
a
n
e
s
ti
mate
of
the
maximum
r
a
dius
a
nd
mi
nim
um
r
a
d
ius
that
the
ir
is
c
a
n
ha
ve
,
the
mi
nim
um
is
c
a
lcula
ted
ba
s
e
d
on
a
pr
e
-
e
s
tablis
he
d.
As
the
f
unc
ti
on
of
the
Hough
tr
a
n
s
f
or
m
r
e
c
e
ives
a
s
a
r
gument
the
im
a
ge
a
nd
the
r
a
dius
,
thi
s
p
r
oc
e
dur
e
is
c
a
r
r
ied
out
withi
n
a
n
it
e
r
a
ti
on
va
r
ying
the
f
r
om
to
a
nd
a
t
the
s
a
me
ti
me
it
is
e
xe
c
uted
twice
s
ince
tw
o
types
o
f
binar
iza
ti
on
a
r
e
us
e
d
.
B
e
c
a
us
e
the
a
lgo
r
it
hm
is
e
xe
c
uted
s
e
ve
r
a
l
ti
mes
a
s
e
t
is
us
e
d
to
s
tor
e
unique
R
OI
s
.
T
he
s
e
t
s
tor
e
s
the
R
OI
s
in
a
r
e
d
-
blac
k
tr
e
e
t
ha
t
will
a
ll
ow
to
a
c
c
e
s
s
the
e
leme
nts
quickly
a
nd
a
voidi
ng
dupli
c
a
tes
.
At
the
e
nd
o
f
thi
s
s
tage
,
ther
e
is
a
s
e
t
of
pos
s
ibl
e
r
e
gions
of
int
e
r
e
s
t,
a
s
s
e
e
n
in
F
igur
e
8.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
4
,
Augus
t
2020
:
2213
-
2223
2218
Algor
it
hm
1
.
I
de
nti
f
y
R
OI
1:
procedure
(
(
,
)
)
2:
←
(
ℎ
(
)
,
ℎ
ℎ
(
)
)
3:
←
/
4:
←
{
}
5:
for
←
do
6: circles.insert(houghTransform(
,
radius))
7:
return
circles
3.
3.
F
il
t
e
r
in
g
of
r
e
gion
s
of
in
t
e
r
e
s
t
As
c
a
n
b
e
s
e
e
n
i
n
F
ig
u
r
e
8
a
la
r
ge
n
um
be
r
o
f
r
e
g
ion
s
of
in
te
r
e
s
t
we
r
e
d
e
t
e
c
ted
,
s
o
me
v
a
l
id
a
nd
o
t
he
r
s
n
o
t
,
s
o
me
wi
t
h
n
ois
e
o
r
ve
r
y
s
im
il
a
r
,
t
hus
i
t
is
n
e
c
e
s
s
a
r
y
a
f
il
te
r
in
g
s
ta
ge
.
R
e
du
c
i
ng
t
he
nu
mb
e
r
o
f
R
O
I
s
r
e
du
c
e
s
t
he
p
r
o
c
e
s
s
in
g
s
tag
e
.
A
lg
o
r
i
t
hm
2
dis
c
a
r
ds
in
va
li
d
c
i
r
c
les
.
S
inc
e
t
he
pe
r
f
o
r
ma
nc
e
o
f
th
e
a
lg
or
i
th
m
de
te
r
io
r
a
tes
u
n
de
r
un
c
o
nt
r
ol
le
d
c
o
nd
it
i
ons
[
1
9
]
,
a
n
d
th
e
r
e
is
a
la
r
ge
n
um
be
r
o
f
s
i
m
il
a
r
R
O
I
s
,
Al
go
r
it
hm
3
wa
s
m
a
d
e
t
o
jo
in
s
im
il
a
r
R
O
I
,
t
his
las
t
one
us
e
s
d
is
jo
in
t
s
e
t
wi
th
u
ni
o
n
by
r
a
nk
a
nd
p
a
t
h
c
o
mp
r
e
s
s
i
on
tha
t
b
y
me
a
ns
o
f
a
δ
o
f
d
is
ta
nc
e
jo
ins
s
i
mi
la
r
R
OI
.
T
he
r
e
s
u
lt
s
o
f
t
his
s
ta
ge
c
a
n
be
s
e
e
n
in
F
i
gu
r
e
9
.
Algor
it
hm
2
.
F
i
lt
e
r
(
R
OI
)
1:
procedure
(
)
2:
←
(img.width*img.height)
3:
←
{
}
4:
for
in
do
5:
←
(
.
)
2
∗
6:
←
/
7:
if
[
,
]
then
8:
if
is
then
9:
validRegions.insert(region)
10:
return
⊳
T
he
s
e
t
w
it
h t
he
v
a
li
d r
e
gi
ons
Algor
it
hm
3
.
J
oin
s
im
il
a
r
(
R
OI
)
1:
procedure
(
)
2:
(
regions.length
)
⊳
I
ns
ta
nc
e
of
S
tr
uc
tu
r
e
U
ni
on F
in
d
3:
for
←
0
to
.
ℎ
do
4:
for
←
+
1
to
.
ℎ
do
5:
if
(
,
)
< delta
then
.
(
,
)
6
:
return
.
_
⊳
T
he
s
e
t
w
it
h a
ll
di
f
f
e
r
e
nt
e
r
e
gi
ons
a
f
te
r
uni
on pr
oc
e
s
s
F
igur
e
8.
P
os
s
ibl
e
r
e
gions
of
int
e
r
e
s
t
de
tec
ted
us
in
g
Algor
it
hm
1
,
obtaine
d
f
r
om
the
dil
a
ted
im
a
ge
s
F
igur
e
9.
R
e
gions
of
I
nter
e
s
t
f
i
lt
e
r
e
d
us
ing
Algor
it
hms
2
a
nd
3
3.
4.
T
r
an
s
f
or
m
a
t
ion
o
f
t
h
e
r
e
gion
o
f
i
n
t
e
r
e
s
t
W
i
th
t
he
de
t
e
c
ti
on
of
th
e
R
O
I
i
t
is
a
p
pl
ie
d
a
t
r
a
ns
f
o
r
m
a
ti
on
tha
t
r
e
d
uc
e
s
no
is
e
in
or
de
r
to
i
s
o
la
te
t
he
i
r
is
a
nd
to
ob
ta
in
t
he
f
e
a
tu
r
e
s
.
T
he
s
c
h
e
me
o
f
F
i
gu
r
e
10
wa
s
us
e
d
t
o
c
a
r
r
y
o
ut
th
e
p
r
o
c
e
s
s
of
un
w
r
a
pp
in
g
t
he
i
r
i
s
.
T
he
ir
is
r
e
gion
is
t
r
a
ns
f
or
med
int
o
a
c
onf
ined
r
e
c
tangula
r
a
r
e
a
,
r
e
c
ognizing
the
bounda
r
ies
is
pos
s
ibl
e
to
a
pply
a
tr
a
ns
f
or
mation
f
r
om
P
olar
c
oor
dinate
s
(
,
)
to
C
a
r
tes
ian
c
oor
dinate
s
(
,
)
,
a
c
c
o
r
ding
to
(
8
)
a
nd
(
9
)
,
whe
r
e
∈
[
0
,
2
]
,
(
,
)
r
e
pr
e
s
e
nts
the
d
ir
e
c
ti
on
of
the
pupil
r
e
gion
whic
h
is
be
ing
tr
a
ns
f
or
med
a
nd
(
,
)
is
the
ne
w
loca
ti
on
of
that
i
r
is
e
leme
nt.
T
he
r
e
s
ul
ts
s
howe
d
in
F
igu
r
e
s
11
a
nd
12
dis
play
the
f
il
ter
e
d
im
a
ge
a
nd
the
t
r
a
ns
f
or
mation
to
C
a
r
tes
ian
c
oor
din
a
tes
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
tec
ti
on
of
uv
e
al
me
lanoma
us
ing
fuz
z
y
and
ne
ur
al
ne
tw
or
k
s
c
las
s
if
ie
r
s
(
Danie
l
F
.
Santos
)
2219
=
+
co
s
(
)
(
8)
=
+
s
in
(
)
(
9)
F
igur
e
10.
S
c
he
me
us
e
d
f
o
r
t
r
a
ns
f
or
m
the
r
e
gion
o
f
int
e
r
e
s
t
F
igur
e
11.
R
e
gion
of
int
e
r
e
s
t
is
olate
d
in
P
olar
c
oor
dinate
s
F
igur
e
12.
I
s
olate
d
r
e
gion
of
int
e
r
e
s
t
in
C
a
r
tes
ian
c
oor
dinate
s
3.
5.
F
e
at
u
r
e
e
x
t
r
ac
t
ion
F
or
the
f
e
a
tu
r
e
s
e
xtr
a
c
ti
on,
the
Hu
mom
e
nts
we
r
e
s
e
lec
ted
due
to
their
pr
ope
r
ti
e
s
mentioned
in
de
tail
in
s
e
c
ti
on
2.
4
.
T
he
e
xtr
a
c
ti
on
wa
s
pe
r
f
or
med
in
a
t
otal
of
1622
im
a
ge
s
that
a
r
e
divi
de
d
a
c
c
or
ding
to
T
a
ble
1.
W
it
h
thes
e
im
a
ge
s
,
the
c
ha
r
a
c
ter
is
ti
c
s
we
r
e
e
xtr
a
c
t
e
d
a
nd
s
t
or
e
d
in
a
c
omm
a
-
s
e
pa
r
a
ted
va
lue
(
c
s
v)
f
il
e
,
whic
h
will
be
late
r
us
e
d
to
t
r
a
in
the
c
las
s
if
ier
s
.
T
a
ble
1.
T
he
nu
mber
o
f
im
a
ge
s
f
o
r
f
e
a
tu
r
e
e
xtr
a
c
ti
on
I
ma
ge
N
umbe
r
W
it
hout
U
ve
a
l
M
e
la
noma
1424
W
it
h U
ve
a
l
M
e
la
noma
198
3.
6.
F
u
z
z
y
m
od
e
l
T
h
e
r
a
nge
s
t
ha
t
d
i
f
f
e
r
e
nt
ia
te
he
a
l
th
y
a
nd
un
he
a
l
th
y
i
r
is
s
e
ts
a
r
e
e
s
ta
b
li
s
he
d
us
in
g
the
da
ta
s
e
t
(
c
s
v
-
f
il
e
)
;
w
i
th
th
is
p
r
o
c
e
s
s
in
g
,
the
me
mbe
r
s
hi
p
f
u
nc
ti
ons
w
e
r
e
bu
il
t
i
n
a
M
a
md
a
n
i
f
uz
z
y
s
ys
te
m
(
F
ig
u
r
e
1
3
)
.
C
on
s
i
de
r
i
ng
t
he
s
e
ve
n
Hu
m
om
e
n
ts
ℎ
1
,
ℎ
2
,
ℎ
3
,
ℎ
4
,
ℎ
5
,
ℎ
6
,
ℎ
7
f
o
r
e
a
c
h
m
om
e
n
t
it
w
a
s
d
e
c
id
e
d
t
o
us
e
t
r
ia
ng
le
f
un
c
t
io
ns
t
o
i
m
pl
e
m
e
n
t
t
he
a
nt
e
c
e
de
n
ts
s
how
n
i
n
F
i
gu
r
e
13
,
w
hi
l
e
Ga
us
s
i
a
n
f
u
nc
ti
ons
a
r
e
i
mp
le
me
nt
e
d
f
o
r
the
c
o
ns
e
qu
e
n
t;
t
h
is
r
e
p
r
e
s
e
n
ts
a
n
u
me
r
i
c
a
l
va
lu
e
t
ha
t
id
e
n
ti
f
ies
i
f
t
he
im
a
g
e
is
h
e
a
lt
hy
or
un
he
a
l
th
y
,
a
s
s
h
ow
n
i
n
F
ig
u
r
e
1
4
.
T
h
e
r
e
s
u
l
t
o
f
th
is
m
od
e
l
c
a
n
be
s
e
e
n
i
n
T
a
b
le
2
i
n
wh
ic
h
it
wa
s
o
bt
a
i
ne
d
7
6
%
o
f
c
o
r
r
e
c
t
c
las
s
if
ic
a
t
io
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
4
,
Augus
t
2020
:
2213
-
2223
2220
F
igur
e
13.
B
lock
diagr
a
m
of
the
f
uz
z
y
inf
e
r
e
nc
e
s
y
s
tem
F
igur
e
14.
F
uz
z
y
s
e
ts
us
e
d
in
the
output
f
or
he
a
lt
hy
or
unhe
a
lt
hy
c
las
s
if
ica
ti
on
3.
7.
Ne
u
r
al
n
e
t
wor
k
s
B
e
c
a
us
e
the
pr
opos
e
d
f
uz
z
y
s
y
s
tem
ha
s
a
s
uc
c
e
s
s
r
a
te
of
76%
,
i
t
wa
s
de
c
ided
to
pe
r
f
or
m
a
ne
ur
a
l
ne
twor
k
c
las
s
if
ier
.
S
uppor
ted
by
a
n
e
xpe
r
im
e
ntal
de
s
ign;
dif
f
e
r
e
nt
c
on
f
igur
a
ti
ons
of
ne
ur
a
l
ne
twor
ks
we
r
e
im
plem
e
nted
to
ob
tain
the
a
c
c
ur
a
c
y
of
the
c
las
s
if
ie
r
.
T
he
ne
u
r
a
l
ne
twor
ks
pa
r
a
mete
r
s
that
a
r
e
c
ha
nge
d
c
a
n
be
s
e
e
n
in
T
a
ble
2.
T
he
inpu
ts
of
the
ne
ur
a
l
ne
twor
k
a
r
e
the
s
e
ve
n
Hu
mom
e
nts
ℎ
1
,
ℎ
2
,
ℎ
3
,
ℎ
4
,
ℎ
5
,
ℎ
6
,
ℎ
7
a
nd
the
output
s
a
r
e
two:
he
a
lt
hy
a
nd
unhe
a
lt
hy
.
F
igu
r
e
1
5
s
hows
a
s
a
mpl
e
c
onf
igu
r
a
ti
on
wi
th
7
input
s
,
2
hidden
laye
r
s
with
3
ne
ur
ons
e
a
c
h
one
a
nd
2
ou
tput
s
.
T
a
ble
2.
P
a
r
a
mete
r
s
a
nd
r
a
nge
s
that
va
r
ied
f
or
the
ne
ur
a
l
ne
twor
k
tes
t
P
a
r
a
ma
te
r
R
a
nge
N
umbe
r
H
id
de
n L
a
ye
r
s
[
1, 10]
N
umbe
r
N
e
ur
ons
pe
r
l
a
ye
r
[
1, 10]
T
ype
of
ne
twor
k
F
e
e
d f
or
w
a
r
d, C
a
s
c
a
de
f
or
w
a
r
d, F
it
ne
t
F
igur
e
15.
S
a
mpl
e
c
on
f
igur
a
ti
on
ne
ur
a
l
ne
twor
k
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
De
tec
ti
on
of
uv
e
al
me
lanoma
us
ing
fuz
z
y
and
ne
ur
al
ne
tw
or
k
s
c
las
s
if
ie
r
s
(
Danie
l
F
.
Santos
)
2221
B
e
low
a
r
e
the
r
e
s
ult
s
with
thr
e
e
types
of
ne
ur
a
l
ne
twor
ks
,
e
a
c
h
c
onf
igur
a
ti
on
wa
s
e
xe
c
uted
f
our
ti
mes
,
the
r
e
s
ult
s
of
one
of
the
e
xe
c
uti
ons
a
r
e
s
hown
be
low;
f
inally,
s
tatis
ti
c
a
l
mea
s
ur
e
s
a
r
e
s
hown
e
a
c
h
c
onf
igur
a
ti
on.
F
igu
r
e
16
r
e
pr
e
s
e
nts
the
mea
n
s
qua
r
e
d
e
r
r
or
(
M
S
E
)
f
or
e
a
c
h
type
of
ne
ur
a
l
ne
twor
k
,
va
r
ying
the
number
of
laye
r
s
a
nd
the
number
of
ne
ur
on
s
ba
s
e
d
on
the
pa
r
a
mete
r
s
of
T
a
ble
2.
T
he
s
e
gr
a
phs
a
nd
T
a
ble
3
pr
e
s
e
nt
the
c
ondit
ions
in
whic
h
the
UM
i
s
mos
t
li
ke
ly
to
be
de
tec
ted.
T
he
c
onf
igur
a
ti
ons
with
low
M
S
E
outcome
s
a
r
e
the
one
s
with
les
s
than
6
ne
ur
o
ns
pe
r
laye
r
a
nd
les
s
than
5
laye
r
s
.
(
a
)
(
b)
(
c
)
F
igur
e
16.
Ne
ur
a
l
ne
two
r
ks
r
e
s
ult
s
(
a
)
pe
r
f
or
manc
e
f
e
e
d
f
or
wa
r
d
n
e
t
,
(
b)
pe
r
f
or
manc
e
c
a
s
c
a
de
f
or
wa
r
d
ne
t
,
(
c
)
pe
r
f
or
ma
nc
e
F
it
Ne
t
T
a
ble
3.
R
e
s
ult
s
of
ne
ur
a
l
ne
two
r
k
c
onf
igu
r
a
ti
ons
T
ype
of
N
e
twor
k
R
e
s
ul
ts
F
e
e
d F
or
w
a
r
d N
e
t
L
a
ye
r
s
/Ne
ur
ons
4
6
7
8
4
0.0720
0.0491
0.0757
0.0814
5
0.0606
0.1065
0.07783
0.0732
6
0.0778
0.0756
0.0763
0.0592
8
0.0728
0.0679
0.0696
0.0867
C
a
s
c
a
d
e
F
or
w
a
r
d N
e
t
L
a
ye
r
s
/Ne
ur
ons
2
4
7
8
4
0.0751
0.0667
0.0828
0.0636
6
0.0717
0.0753
0.0819
0.0824
8
0.0712
0.0786
0.0647
0.0606
9
0.08198
0.0611
0.0495
0.0
822
F
it
N
e
t
L
a
ye
r
s
/Ne
ur
ons
4
5
6
8
1
0.0783
0.0780
0.0823
0.0629
2
0.0629
0.0760
0.0750
0.0757
4
0.0775
0.0729
0.0467
0.0843
6
0.0756
0.0630
0.0753
0.0728
4.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
As
a
n
e
xpe
r
im
e
ntal
r
e
s
ult
,
the
a
c
c
ur
a
c
y
r
a
tes
of
the
f
uz
z
y
s
ys
tem
a
nd
the
ne
ur
a
l
ne
twor
k
s
ys
tem
a
r
e
s
hown.
I
n
the
c
onf
us
ion
matr
ix
of
T
a
ble
4
a
r
e
s
hown
the
r
e
s
ult
s
o
f
the
f
uz
z
y
c
las
s
if
ier
,
the
f
unc
ti
on
that
wa
s
us
e
d
to
pe
r
f
or
m
the
c
las
s
if
ica
ti
on
c
a
n
be
s
e
e
n
in
the
(
10
)
whe
r
e
is
the
f
uz
z
y
s
ys
tem
c
r
e
a
ted
,
is
the
f
unc
ti
on
that
obtains
the
va
lues
of
Hu
a
nd
is
the
im
a
ge
to
be
tes
ted.
As
c
a
n
be
s
e
e
n
in
the
c
o
nf
us
ion
matr
ix
of
T
a
ble
4
,
the
f
uz
z
y
s
ys
tem
ge
ne
r
a
tes
be
tt
e
r
r
e
s
ult
s
with
he
a
lt
hy
im
a
ge
s
.
T
he
hypothes
is
of
thes
e
r
e
s
ult
s
is
ba
s
e
d
on
the
f
a
c
t
that
the
e
ntr
y
da
tas
e
t
is
lar
ge
r
f
or
he
a
lt
hy
ir
is
than
f
or
unhe
a
lt
hy
ir
is
.
Give
n
the
las
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
1693
-
6930
T
E
L
KO
M
NI
KA
T
e
lec
omm
un
C
omput
E
l
C
ontr
o
l
,
Vol.
18
,
No
.
4
,
Augus
t
2020
:
2213
-
2223
2222
table
is
pos
s
ibl
e
to
c
omput
e
s
ome
mea
s
ur
e
s
li
ke
a
c
c
ur
a
c
y,
e
r
r
o
r
r
a
te,
s
e
ns
it
ivi
ty
a
nd
pr
e
c
is
ion
that
c
a
n
give
mor
e
ins
ight
s
o
f
the
model,
the
r
e
s
ult
s
a
r
e
0
.
7663
3,
0
.
23366,
0
.
88847,
a
nd
0.
83917
,
r
e
s
pe
c
ti
ve
ly.
R
e
ga
r
ding
the
pr
e
c
is
ion
a
nd
s
e
ns
it
ivi
ty
c
a
n
b
e
s
hown
that
the
s
ys
tem
is
pr
one
to
de
tec
t
he
a
lt
hy
im
a
ge
s
.
Af
ter
e
xe
c
uti
ng
the
dif
f
e
r
e
nt
c
onf
igur
a
ti
ons
of
ne
ur
a
l
ne
twor
ks
,
a
s
tatis
ti
c
a
l
a
na
lys
is
wa
s
pe
r
f
or
med
on
f
our
da
ta
-
s
e
ts
c
or
r
e
s
ponding
to
the
r
e
s
ult
s
of
e
a
c
h
c
onf
igur
a
ti
on;
r
e
pr
e
s
e
nts
the
s
tand
a
r
d
de
viation
of
the
r
e
s
ult
s
of
the
e
xe
c
uti
on
number
.
T
he
r
e
s
ult
s
obtaine
d
a
r
e
s
h
own
in
T
a
ble
5
,
whe
r
e
it
c
a
n
be
s
e
e
n
that
the
ne
tw
or
k
that
pr
oduc
e
s
the
be
s
t
r
e
s
ult
s
is
the
f
e
e
d
f
or
wa
r
d
ne
twor
k
with
a
s
uc
c
e
s
s
r
a
te
of
96
.
04%
.
(
)
=
{
(
(
)
)
>
0
ℎ
ℎ
(
(
)
)
<
0
ℎ
ℎ
(
10)
T
a
ble
4.
C
onf
us
ion
matr
ix
us
ing
the
p
r
opos
e
d
f
uz
z
y
s
ys
tem
T
r
ue
di
a
gnos
is
H
e
a
lt
hy
U
nhe
a
lt
hy
T
ot
a
l
H
e
a
lt
hy
1195
229
1424
U
nhe
a
lt
hy
150
48
198
T
ot
a
l
1345
277
1622
T
a
ble
5.
S
tanda
r
d
de
viation
a
nd
M
S
E
T
ype
of
N
e
twor
k
1
1
2
2
3
3
4
4
F
e
e
d F
or
w
a
r
d
0.0093
0.0491
0.4598
0.0408
0.0080
0.0396
0.0098
0.0442
C
a
s
c
a
d
e
F
or
w
a
r
d
0.0066
0.0495
0.00110
0.0463
0.0095
0.0431
0.0238
0.0466
F
it
N
e
t
0.0090
0.0467
0.0205
0.0445
0.0107
0.0446
0.0066
0.0432
F
i
n
a
l
l
y
,
j
u
d
g
i
n
g
f
r
o
m
t
h
e
r
e
s
u
l
t
s
c
o
l
l
e
c
t
e
d
t
h
r
o
u
g
h
t
h
e
d
i
f
f
e
r
e
n
t
e
x
e
c
u
t
i
o
n
s
,
i
t
c
a
n
b
e
s
e
e
n
t
h
a
t
t
h
e
f
e
e
d
f
o
r
w
a
r
d
n
e
t
w
o
r
k
p
r
o
d
u
c
e
s
t
h
e
b
e
s
t
a
c
c
u
r
a
c
y
o
f
96
.
04%
w
i
t
h
s
t
a
n
d
a
r
d
d
e
v
i
a
t
i
o
n
o
f
0
.
008
w
h
i
c
h
c
a
n
b
e
i
m
p
r
o
v
e
d
f
r
o
m
d
i
f
f
e
r
e
n
t
a
p
p
r
o
a
c
h
e
s
s
u
c
h
a
s
i
n
c
r
e
a
s
i
n
g
t
h
e
s
i
z
e
o
f
t
h
e
d
a
t
a
s
e
t
,
i
n
c
r
e
m
e
n
t
i
n
g
t
h
e
n
u
m
b
e
r
o
f
d
e
s
c
r
i
p
t
o
r
s
,
e
.
g
G
a
b
o
r
d
e
s
c
r
i
p
t
o
r
o
r
u
s
i
n
g
o
t
h
e
r
t
e
c
h
n
i
q
u
e
s
,
e
.
g
A
N
F
I
S
,
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
s
o
r
g
e
n
e
t
i
c
a
l
g
o
r
i
t
h
m
s
.
5.
CONC
L
USI
ON
T
he
pr
opos
e
d
methodology
wa
s
tes
ted
us
ing
di
f
f
e
r
e
nt
c
onf
igu
r
a
ti
ons
,
the
e
xpe
r
im
e
ntal
r
e
s
ult
s
s
how
that
f
uz
z
y
logi
c
a
nd
ne
ur
a
l
ne
twor
ks
c
las
s
if
ier
s
pr
ovide
a
s
uit
a
ble
s
ys
tem
to
de
tec
t
uve
a
l
m
e
lanoma
,
a
c
hieving
76%
in
the
f
uz
z
y
c
las
s
if
ier
a
nd
f
or
the
ne
u
r
a
l
ne
twor
k
c
las
s
if
ier
whic
h
pe
r
f
or
ms
be
tt
e
r
with
a
n
a
c
c
ur
a
c
y
of
96
.
04%
us
ing
a
F
e
e
d
F
o
r
wa
r
d
Ne
t
.
F
o
r
the
s
e
gmenta
ti
on
of
the
R
OI
i
t
wa
s
pr
opos
e
d
a
ne
w
a
lgor
it
hm
buil
t
ove
r
the
pr
inciples
o
f
W
il
de
s
a
lgor
it
hm,
th
is
ne
w
a
lgor
it
hm
wa
s
done
to
im
pr
ove
the
s
uc
c
e
s
s
r
a
ti
o
o
f
we
ll
-
s
e
gmente
d
r
e
gions
of
int
e
r
e
s
t
(
R
OI
s
)
a
s
we
ll
a
s
the
in
tegr
a
ti
on
of
da
ta
s
tr
uc
tur
e
s
li
ke
dis
joi
nt
-
s
e
t
with
pa
th
a
nd
r
a
nge
c
ompr
e
s
s
ion
to
r
e
duc
e
the
pr
o
c
e
s
s
ing
ti
me.
As
a
r
e
s
ult
of
the
pr
e
-
pr
oc
e
s
s
ing
s
tag
e
,
it
wa
s
pos
s
ibl
e
to
pe
r
f
o
r
m
de
tec
ti
on
o
f
the
r
e
gions
of
int
e
r
e
s
t.
T
he
s
tage
o
f
tr
a
ns
f
or
mation
a
nd
u
s
e
of
da
ta
s
t
r
uc
tur
e
s
wa
s
de
c
is
ive
to
r
e
duc
e
the
nois
e
in
the
da
ta
s
e
t
a
n
d
the
p
r
oc
e
s
s
ing
ti
me
f
o
r
the
f
e
a
tur
e
e
xtr
a
c
ti
on
.
I
n
or
de
r
to
im
pr
ove
the
pe
r
f
or
manc
e
of
the
c
las
s
if
ier
s
,
it
is
c
ons
ider
e
d
to
e
xpa
nd
the
number
of
im
a
ge
s
us
e
d,
a
s
we
ll
a
s
the
int
e
gr
a
t
ion
of
mor
e
de
s
c
r
ipt
o
r
s
int
o
the
t
r
a
ini
ng
da
ta
s
e
t
s
uc
h
a
s
G
a
bor
de
s
c
r
ipt
or
s
.
I
n
a
f
utu
r
e
wor
k
the
methodology
a
nd
the
a
lgor
it
hm
pr
opos
e
d
c
ould
be
im
pleme
nted
in
s
mar
thphones
AC
KNOWL
E
DGM
E
N
T
T
he
a
uthor
s
e
xpr
e
s
s
gr
a
ti
tude
to
the
F
a
c
ult
a
d
de
I
nge
nier
ía
of
the
Unive
r
s
idad
Dis
tr
i
tal
F
r
a
nc
is
c
o
J
os
é
de
C
a
ldas
,
a
nd
a
ls
o
to
the
Dr
.
P
a
ul
T
.
F
inger
,
M
D.
RE
F
E
RE
NC
E
S
[
1
]
K
.
Ma
h
e
n
d
r
ar
a
j
,
S.
S
h
re
s
t
h
a,
C
.
S.
M.
L
a
u
,
a
n
d
R.
S.
C
h
a
m
b
er
l
a
i
n
,
“
O
c
u
l
a
r
m
e
l
a
n
o
ma
-
w
h
e
n
y
o
u
h
a
v
e
s
e
e
n
o
n
e,
y
o
u
h
a
v
e
n
o
t
s
e
e
n
t
h
em
a
l
l
:
a
c
l
i
n
i
ca
l
o
u
t
c
o
m
e
s
t
u
d
y
fr
o
m
t
h
e
s
u
r
v
e
i
l
l
a
n
c
e,
e
p
i
d
em
i
o
l
o
g
y
a
n
d
e
n
d
r
e
s
u
l
t
s
(
s
ee
r)
d
a
t
a
b
a
s
e
(
1
9
7
3
-
2
0
1
2
),
”
C
l
i
n
i
c
a
l
O
p
h
t
h
a
l
m
o
l
o
g
y
,
v
o
l
.
2017
,
n
o
.
11
,
p
p
.
153
-
160
,
2
0
1
7
.
[
2
]
N
a
t
i
o
n
a
l
O
r
g
a
n
i
za
t
i
o
n
f
o
r
Ra
re
D
i
s
o
r
d
e
r
s
,
“R
ar
e
d
i
s
e
a
s
e
d
a
t
a
b
a
s
e,
”
2
0
1
9
.
[
O
n
l
i
n
e
].
A
v
a
i
l
a
b
l
e
:
https://
r
a
r
e
dis
e
a
s
e
s.
or
g/
r
a
re
-
d
i
s
e
a
s
e
s
/
o
c
u
l
a
r
-
me
l
a
n
o
m
a
/
.
A
cc
e
s
s
e
d
:
S
e
p
t
e
m
b
e
r
2
0
1
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.