T
E
L
KO
M
NIK
A
, V
ol
.
17
,
No.
5,
O
c
tob
er
20
1
9,
p
p.
2
3
46
~
2
3
54
IS
S
N: 1
69
3
-
6
93
0
,
accr
ed
ited
F
irst
Gr
ad
e b
y K
em
en
r
istekdikti,
Decr
ee
No:
21
/E
/
K
P
T
/20
18
DOI:
10.12928/TE
LK
OM
N
IK
A
.v
1
7
i
5
.
12551
◼
23
46
Rec
ei
v
ed
M
arc
h
12
,
20
1
9
;
Rev
i
s
ed
A
pri
l
1
1
, 2
01
9
; A
c
c
ep
ted
Ma
y
9
,
20
1
9
Effect
i
v
e se
gm
e
nt
ati
on
of
s
cler
a, iri
s an
d p
u
pil
in n
oi
s
y
e
y
e i
m
ages
M
r
u
n
al
P
ath
ak
*
1
,
N.
S
r
iniv
as
u
2
,
V
ina
ya
k B
air
agi
3
1
,2
K.
L.
U
n
i
v
e
rs
i
ty
,
In
d
i
a
3
AISSM
S I
n
s
t.
o
f
In
fo
.
T
e
c
h
.,
S.
P.
Pu
n
e
Un
i
v
e
rs
i
ty
,
In
d
i
a
*C
o
rre
s
p
o
n
d
i
n
g
a
u
th
o
r,
e
-
m
a
i
l
:
m
ru
n
a
l
k
p
a
th
a
k
@g
m
a
i
l
.c
o
m
Ab
strac
t
In
t
o
d
a
y
’
s
s
e
n
s
i
t
i
v
e
e
n
v
i
r
o
n
m
e
n
t,
fo
r
p
e
r
s
o
n
a
l
a
u
th
e
n
t
i
c
a
t
i
o
n
,
i
r
i
s
re
c
o
g
n
i
ti
o
n
i
s
th
e
m
o
s
t
a
tt
e
n
ti
v
e
te
c
h
n
i
q
u
e
a
m
o
n
g
t
h
e
v
a
r
i
o
u
s
b
i
o
m
e
tri
c
t
e
c
h
n
o
l
o
g
i
e
s
.
O
n
e
o
f
th
e
k
e
y
s
te
p
s
i
n
th
e
i
ri
s
r
e
c
o
g
n
i
ti
o
n
s
y
s
te
m
i
s
th
e
a
c
c
u
ra
t
e
i
ri
s
s
e
g
m
e
n
t
a
ti
o
n
fr
o
m
i
t
s
s
u
rro
u
n
d
i
n
g
n
o
i
s
e
s
i
n
c
l
u
d
i
n
g
p
u
p
i
l
a
n
d
s
c
l
e
r
a
o
f
a
c
a
p
tu
r
e
d
e
y
e
-
i
m
a
g
e
.
In
o
u
r
p
ro
p
o
s
e
d
m
e
th
o
d
,
i
n
i
ti
a
l
l
y
i
n
p
u
t
i
m
a
g
e
i
s
p
re
p
r
o
c
e
s
s
e
d
b
y
u
s
i
n
g
b
i
l
a
te
r
a
l
fi
l
te
r
i
n
g
.
Aft
e
r
th
e
p
re
p
ro
c
e
s
s
i
n
g
o
f
i
m
a
g
e
s
c
o
n
to
u
r
b
a
s
e
d
fe
a
t
u
re
s
s
u
c
h
a
s
,
b
ri
g
h
tn
e
s
s
,
c
o
l
o
r
a
n
d
te
x
tu
r
e
fe
a
t
u
re
s
a
re
e
x
tr
a
c
t
e
d
.
T
h
e
n
e
n
tro
p
y
i
s
m
e
a
s
u
r
e
d
b
a
s
e
d
o
n
th
e
e
x
tra
c
te
d
c
o
n
to
u
r
b
a
s
e
d
fe
a
tu
re
s
to
e
ff
e
c
t
i
v
e
l
y
d
i
s
t
i
n
g
u
i
s
h
i
n
g
th
e
d
a
ta
i
n
t
h
e
i
m
a
g
e
s
.
F
i
n
a
l
l
y
,
th
e
c
o
n
v
o
l
u
ti
o
n
n
e
u
ra
l
n
e
two
rk
(CN
N)
i
s
u
s
e
d
fo
r
th
e
e
ff
e
c
ti
v
e
s
c
l
e
ra
,
i
r
i
s
a
n
d
p
u
p
i
l
p
a
rts
s
e
g
m
e
n
t
a
ti
o
n
s
b
a
s
e
d
o
n
th
e
e
n
tro
p
y
m
e
a
s
u
r
e
.
Th
e
p
ro
p
o
s
e
d
re
s
u
l
ts
a
re
a
n
a
l
y
z
e
d
to
d
e
m
o
n
s
tra
t
e
th
e
b
e
tt
e
r
p
e
r
fo
rm
a
n
c
e
o
f
th
e
p
ro
p
o
s
e
d
s
e
g
m
e
n
ta
t
i
o
n
m
e
th
o
d
th
a
n
th
e
e
x
i
s
ti
n
g
m
e
th
o
d
s
.
Key
w
ords
:
c
o
n
to
u
r
-
b
a
s
e
d
fe
a
tu
re
s
,
e
n
tr
o
p
y
m
e
a
s
u
r
e
, n
o
rm
a
l
i
z
a
ti
o
n
,
p
r
e
p
ro
c
e
s
s
i
n
g
,
s
e
g
m
e
n
ta
t
i
o
n
Copy
righ
t
©
2
0
1
9
Uni
v
e
rsi
t
a
s
Ahm
a
d
D
a
hl
a
n.
All
rig
ht
s
r
e
s
e
rve
d
.
1.
Int
r
o
d
u
ctio
n
Ir
i
s
r
ec
og
n
i
ti
on
em
ergi
n
g
a
s
on
e
of
th
e
m
os
t
pref
err
e
d
bi
om
etri
c
tec
hn
o
l
og
y
m
od
al
i
ti
es
f
or
au
tom
ate
d
pe
r
s
on
al
i
d
e
nti
f
i
c
ati
on
[1
-
3].
It
i
s
a
b
i
o
m
etri
c
r
ec
og
ni
ti
o
n
tec
h
no
l
o
g
y
t
ha
t
uti
l
i
z
es
pa
tte
r
n
r
ec
og
n
i
t
i
on
t
ec
hn
i
qu
es
on
the
ba
s
i
s
of
i
r
i
s
hi
gh
qu
a
l
i
t
y
i
m
ag
es
[4
,
5].
S
i
nc
e
i
n
c
o
m
pa
r
i
s
on
wi
th
o
the
r
f
ea
tures
,
i
r
i
s
r
ec
og
n
i
t
i
on
i
s
be
s
t
bi
om
etri
c
tec
h
no
l
og
i
es
[6]
.
Ir
i
s
s
eg
m
en
tat
i
on
u
nd
er
v
i
s
i
b
l
e
s
pe
c
tr
um
(
V
IS
)
i
s
s
ti
l
l
a
v
er
y
c
h
al
l
en
gi
ng
prob
l
em
.
Non
c
o
op
era
ti
v
e
i
r
i
s
r
ec
og
n
i
ti
on
r
ef
ers
to
au
tom
ati
c
al
l
y
r
ec
og
n
i
z
e
at
a
di
s
tan
c
e
an
d
d
ea
l
i
n
g
w
i
t
h
s
ev
era
l
f
ac
tor
s
tha
t
de
t
erio
r
ate
the
q
ua
l
i
t
y
of
an
i
m
ag
e
[7
].
Ma
n
y
a
l
go
r
i
thm
s
ha
v
e
b
ee
n
pro
po
s
e
d
f
or
s
ep
arati
ng
the
i
r
i
s
r
eg
i
on
f
r
om
the
no
n
-
i
r
i
s
r
eg
i
on
s
o
n
i
m
ag
es
.
O
ne
of
the
m
ai
n
ap
proac
h
es
c
on
s
i
s
ts
on
bo
un
da
r
y
-
ba
s
ed
m
eth
od
s
[
8
-
11]
.
Mo
r
eo
v
er,
s
eg
m
en
tat
i
on
of
the
s
c
l
era
r
eg
i
o
n
he
l
ps
to
i
m
prov
e
i
r
i
s
r
ec
og
ni
t
i
on
ac
c
urac
y
un
de
r
di
f
f
erent
l
i
gh
t
i
ng
c
o
nd
i
t
i
o
ns
an
d
e
y
e
ga
z
es
[
12
].
A
f
ter
de
c
ad
es
of
r
es
ea
r
c
h
on
i
r
i
s
r
ec
og
ni
t
i
o
n,
t
he
t
ec
hn
o
l
o
g
y
i
s
no
w
h
ea
d
i
ng
to
i
m
prov
e
r
ec
og
n
i
t
i
on
pe
r
f
orm
an
c
e
b
y
c
om
bi
ni
ng
oth
er
oc
ul
ar
(
r
eti
n
a,
c
or
ne
a,
s
c
l
era,
pu
p
i
l
,
pe
r
i
o
c
ul
ar)
or
n
on
-
oc
u
l
ar
(
f
ac
e,
f
i
ng
erpr
i
nt
,
pa
l
m
prin
t
etc
.)
m
od
al
i
t
i
e
s
[
1
3
].
A
l
t
ho
u
gh
th
e
ac
c
urac
i
es
of
the
v
i
s
i
bl
e
s
pe
c
tr
um
,
i
r
i
s
r
ec
og
n
i
ti
on
s
y
s
t
em
s
are
no
t
c
om
pa
r
ab
l
e
t
o
th
os
e
o
pe
r
at
i
ng
i
n
th
e
ne
ar
i
nf
r
ared
s
p
ec
tr
um
[
1
4
],
t
he
v
i
s
i
b
l
e
s
pe
c
tr
um
i
r
i
s
i
m
ag
i
ng
h
as
the
ad
v
a
nta
g
e
of
pe
r
m
i
tti
n
g
th
e
i
nt
eg
r
at
i
o
n
of
ad
d
i
t
i
o
na
l
s
ou
r
c
es
of
i
nf
orm
ati
on
, s
uc
h a
s
e
y
e c
o
l
or or
s
c
l
era
v
as
c
ul
at
ure [1
5
].
A
m
eth
od
f
or
s
c
l
era
s
e
g
m
en
tat
i
on
ba
s
e
d
o
n
F
u
z
z
y
l
og
i
c
i
s
pro
po
s
ed
b
y
[1
6,
1
7
].
S
V
M
an
d
f
ea
t
ure
s
el
ec
t
i
on
tec
hn
i
q
ue
s
[1
8
],
C
i
r
c
ul
ar
Ho
ug
h
T
r
an
s
f
or
m
a
nd
K
-
M
ea
ns
al
g
orit
hm
[
19
]
Re
v
ers
e
b
i
o
-
w
a
v
e
l
et
tr
an
s
f
orm
[20]
f
or
i
r
i
s
r
ec
o
gn
i
ti
o
n.
T
he
h
i
g
h
es
t
di
f
f
i
c
ul
t
y
of
hu
m
an
i
r
i
s
s
eg
m
en
tat
i
on
i
s
tha
t
i
t
i
s
ha
r
d
to
di
s
c
o
v
er
t
he
a
pp
are
nt
f
ea
tur
e
v
al
ue
s
i
n
t
he
i
m
ag
e
an
d
to
k
ee
p
t
he
i
r
r
ep
r
es
e
n
t
c
ap
a
bi
l
i
t
y
hi
g
h
i
n
a
prof
i
c
i
en
t
m
an
ne
r
[
21
].
A
l
s
o,
d
i
f
fi
c
ul
t
y
i
n
s
c
l
era
s
eg
m
en
tat
i
on
ar
i
s
es
f
r
o
m
the
i
nc
l
us
i
o
n
of
e
y
e
l
i
ds
a
nd
e
y
el
as
he
s
i
n
th
e
s
c
l
era
r
eg
i
on
an
d
t
he
no
ti
c
e
ab
l
e
ef
f
ec
t
of
l
i
gh
t
i
n
g
c
on
di
t
i
on
s
.
T
he
pe
r
f
or
m
an
c
e
of
i
r
i
s
,
s
c
l
era
an
d
r
ec
og
ni
t
i
on
s
y
s
tem
s
hi
g
hl
y
d
ep
e
nd
s
o
n t
h
e s
eg
m
en
tat
i
on
pr
oc
es
s
whi
c
h
i
s
a
c
ha
l
l
e
ng
i
ng
prob
l
em
.
T
he
m
ai
n
c
on
tr
i
b
uti
on
s
of
thi
s
pa
pe
r
ar
e
as
f
ol
l
o
w
s
:
1)
c
on
tou
r
B
as
e
d
F
e
atu
r
es
are
ex
tr
ac
ted
f
or
t
he
ef
f
ec
ti
v
e
s
eg
m
en
tat
i
on
of
pu
p
i
l
,
i
r
i
s
an
d
s
c
l
era
;
2)
e
ntro
p
y
i
s
m
ea
s
ured
to
ef
f
ec
ti
v
el
y
di
s
t
i
n
gu
i
s
h
i
n
g
th
e
da
t
a
i
n
the
i
m
ag
es
;
3)
t
h
e
CNN
b
as
ed
on
th
e
e
ntro
p
y
m
ea
s
ure
i
s
us
ed
f
or the
ef
f
ec
ti
v
e s
c
l
era
, i
r
i
s
a
nd
p
up
i
l
pa
r
ts
s
eg
m
en
tat
i
on
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
E
ffe
c
t
i
v
e s
eg
me
nt
ati
on
of
s
c
l
era, i
r
i
s
an
d p
u
pi
l
i
n n
oi
s
y
ey
e i
ma
g
es
(
M
r
un
a
l
P
at
ha
k
)
2347
2.
Rel
ated
W
o
r
k
A
nd
r
e
a
F
.
A
ba
t
e
et
a
l
.
[
2
2
]
propos
e
d
a
n
ov
el
h
um
an
i
r
i
s
r
ec
og
ni
t
i
on
ap
pro
ac
h
b
as
ed
on
a
m
ul
ti
-
l
a
y
er
p
erc
ep
tr
on
NN
an
d
pa
r
t
i
c
l
e
s
war
m
op
ti
m
i
z
at
i
o
n
(
P
S
O
)
al
go
r
i
t
hm
s
to
tr
ai
n
the
ne
t
wor
k
i
n
order
to
i
nc
r
ea
s
e
ge
ne
r
a
l
i
z
at
i
on
pe
r
f
orm
an
c
e.
A
c
o
m
bi
na
ti
o
n
of
the
s
e
al
go
r
i
t
hm
s
was
us
ed
as
a
c
l
as
s
i
f
i
er.
A
P
S
O
a
l
go
r
i
thm
was
ap
pl
i
ed
to
tr
a
i
n
the
NN
f
or
da
t
a
c
l
as
s
i
f
i
c
at
i
on
.
Nagl
aa
et
al
.
[2
3
]
pres
e
nte
d
a
c
o
ars
e
-
to
-
f
i
ne
al
g
orit
hm
f
or
eff
i
c
i
en
t
Ir
i
s
L
oc
al
i
z
a
ti
o
n
an
d
Rec
og
n
i
ti
on
,
wh
i
l
e
ac
hi
ev
i
n
g
an
ac
c
ep
ta
bl
e
ac
c
urac
y
.
T
he
i
r
i
s
gra
y
i
m
ag
e
w
as
tr
a
ns
f
or
m
ed
to
a
bi
n
ar
y
i
m
ag
e
us
i
ng
an
ad
ap
ti
v
e
thres
h
ol
d
o
bta
i
n
ed
f
r
om
an
al
y
z
i
n
g
t
he
i
m
ag
e
i
nte
ns
i
t
y
hi
s
tog
r
am
.
F
i
na
l
l
y
,
a
r
ef
i
n
em
en
t
s
tep
w
as
m
ad
e
us
i
ng
an
i
nte
gral
-
d
i
f
f
erenti
a
l
op
erat
or
to
ge
t
the
f
i
na
l
i
r
i
s
an
d p
up
i
l
c
en
te
r
s
.
Mo
ha
m
m
ed
A
.
M.
A
b
du
l
l
ah
e
t
a
l
.
[2
4
]
pro
po
s
ed
a
no
v
e
l
s
eg
m
en
tat
i
on
m
eth
od
f
or
non
-
i
de
a
l
i
r
i
s
i
m
ag
es
.
T
w
o
al
g
orit
hm
s
w
ere
prop
os
ed
f
or
pu
p
i
l
s
eg
m
en
tat
i
on
i
n
i
r
i
s
i
m
ag
es
,
the
y
wer
e
c
a
ptu
r
e
d
un
d
er
v
i
s
i
bl
e
a
nd
ne
ar
i
nf
r
ared
l
i
gh
t.
T
he
propos
e
d
s
c
he
m
e
w
as
r
ob
us
t
i
n
f
i
nd
i
n
g
the
ex
ac
t
i
r
i
s
bo
u
nd
ar
y
a
nd
i
s
ol
at
i
n
g
the
e
y
e
l
i
ds
of
the
i
r
i
s
i
m
ag
es
.
A
l
k
as
s
aret
al
.
[2
5
]
pres
en
ted
the
de
s
i
gn
of
a
r
ob
us
t
s
c
l
era
r
ec
og
n
i
ti
on
s
y
s
tem
w
i
t
h
hi
g
h
ac
c
urac
y
.
T
he
y
al
s
o
propos
e
d
an
ef
f
i
c
i
en
t
m
eth
od
f
or
v
es
s
el
en
ha
nc
em
en
t,
ex
tr
ac
ti
on
,
a
nd
bi
na
r
i
z
a
ti
o
n.
In
t
he
f
ea
t
u
r
e
ex
tr
ac
ti
o
n
an
d
m
atc
hi
ng
proc
es
s
s
tag
es
,
the
y
ad
di
t
i
o
na
l
l
y
d
ev
e
l
o
pe
d
a
n
ef
f
i
c
i
en
t
m
e
tho
d,
t
ha
t
i
s
,
orie
nt
ati
on
,
s
c
al
e,
i
l
l
um
i
n
ati
o
n,
an
d
de
f
orm
ati
on
i
nv
ar
i
an
t
.
P
at
tab
hi
Ram
ai
ah
an
d
A
j
a
y
K
um
ar
[2
6
]
ha
v
e
de
v
el
op
ed
a
d
om
ai
n
ad
a
pta
t
i
o
n
f
r
am
ew
ork
to
ad
dres
s
the
probl
em
an
d
i
ntrod
uc
ed
a
n
e
w
a
l
go
r
i
th
m
us
i
ng
Ma
r
k
ov
r
an
do
m
f
i
el
ds
(
MRF)
m
od
el
to
s
i
gn
i
f
i
c
an
t
l
y
i
m
prov
e
c
r
os
s
-
do
m
ai
n i
r
i
s
r
ec
og
ni
t
i
o
n.
3.
Re
se
a
r
ch M
eth
o
d
T
hi
s
pa
pe
r
pres
en
ts
a
pr
of
i
c
i
en
t
s
eg
m
en
tat
i
on
of
i
r
i
s
,
s
c
l
era,
an
d
pu
p
i
l
uti
l
i
z
i
ng
ef
f
ec
ti
v
e
f
ea
tures
an
d
CN
N
c
l
us
ter
i
n
g.
Her
e,
t
he
CN
N
s
uc
c
es
s
ful
l
y
c
l
us
ters
the
d
ata
i
n
i
m
ag
es
ba
s
e
d
on
the
s
i
m
i
l
ari
t
y
ob
t
ai
ne
d
b
y
the
e
ntrop
y
m
ea
s
ure
a
nd
s
ub
s
eq
ue
ntl
y
r
es
ul
t
s
t
he
s
c
l
era,
i
r
i
s
an
d
pu
p
i
l
s
e
gm
en
ts
s
ep
arate
l
y
.
T
he
proc
es
s
i
ng
f
l
o
w
of
th
e
propos
e
d s
tr
ate
g
y
i
s
gi
v
en
i
n
F
i
g
ure 1
.
F
i
gu
r
e
1.
P
r
oc
es
s
i
ng
f
l
o
w o
f
propo
s
ed
m
eth
od
3.1
.
P
r
eproce
ss
ing
A
t
f
i
r
s
t
i
np
ut
i
m
ag
e
i
s
tak
en
f
r
o
m
da
tab
as
e
a
nd
prep
r
oc
es
s
ed
b
y
us
i
ng
no
r
m
al
i
z
at
i
on
proc
es
s
an
d
b
i
l
ate
r
a
l
f
i
l
te
r
i
ng
to
en
ha
nc
e
t
he
f
urt
he
r
proc
es
s
i
ng
.
T
he
pre
-
proc
es
s
i
ng
is
de
s
c
r
i
be
d
i
n
f
ol
l
o
wi
n
g s
ec
ti
on
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
2
3
46
-
2
3
54
2348
3.1.1
.
No
r
m
ali
z
atio
n
Nor
m
al
i
z
at
i
on
ac
h
i
e
v
es
t
h
e
l
i
ne
ar
tr
a
ns
f
or
m
ati
on
of
the
i
m
ag
e
to
f
i
t
i
nto
a
pa
r
ti
c
ul
ar
r
an
ge
.
Her
e,
M
i
n
-
m
ax
no
r
m
al
i
z
ati
on
proc
ed
ure
i
s
ut
i
l
i
z
ed
f
or
th
e
s
ta
nd
ard
i
z
at
i
on
of
i
m
ag
e
whi
c
h
l
i
ne
ar
l
y
tr
a
ns
f
or
m
s
the
i
nf
or
m
ati
on
.
M
i
n
-
M
ax
no
r
m
al
i
z
at
i
o
n
i
s
d
on
e
thr
ou
g
h
the
ac
c
om
pa
n
y
i
ng
c
on
d
i
ti
o
n (1)
:
m
i
n
m
a
x
m
i
n
~
~
~
~
Y
Y
Y
Y
N
−
−
=
(
1)
w
he
r
e,
m
i
n
~
Y
and
m
a
x
~
Y
are
t
he
m
i
ni
m
u
m
an
d
m
ax
i
m
u
m
v
al
u
es
i
n
i
m
ag
e
Y
~
,
w
h
ere
N
i
s
the
n
orm
al
i
z
e
d i
m
ag
e.
3.1.2
.
Bil
atera
l filt
er
ing
T
he
bi
l
ate
r
al
f
i
l
ter
tak
es
a
w
e
i
g
hte
d
s
um
of
the
p
i
x
el
s
i
n
a
ne
arb
y
ne
i
gh
b
o
r
ho
od
;
the
wei
gh
ts
r
e
l
y
u
po
n
bo
t
h
the
s
pa
t
i
al
d
i
s
tan
c
e
an
d
t
he
i
nte
ns
i
t
y
di
s
ta
nc
e.
P
r
ec
i
s
el
y
,
at
a
pi
x
e
l
l
oc
at
i
on
x
,
the
ou
t
pu
t
of
a
bi
l
ate
r
a
l
f
i
l
t
er i
s
c
om
pu
ted
as
f
ol
l
o
w
s
:
)
(
1
)
(
ˆ
2
2
2
2
2
)
(
)
(
)
(
2
y
I
e
e
C
x
I
r
d
x
I
y
I
x
N
y
x
y
=
−
−
−
−
(
2)
w
he
r
e,
d
an
d
r
are
pa
r
am
ete
r
s
c
on
tr
ol
l
i
n
g
t
he
f
al
l
-
of
f
of
wei
gh
ts
i
n
s
pa
ti
al
an
d
i
nte
ns
i
t
y
do
m
ai
ns
,
i
nd
i
v
i
d
ua
l
l
y
,
)
(
x
N
i
s
a
s
pa
ti
a
l
ne
i
g
hb
orh
oo
d
of
pi
x
el
)
(
x
I
,
an
d
C
i
s
the
no
r
m
al
i
z
ati
on
c
on
s
tan
t.
T
hi
s
B
i
l
ate
r
a
l
f
i
l
t
e
r
i
s
m
os
tl
y
ut
i
l
i
z
ed
f
or
s
m
o
oth
i
ng
th
e
i
m
ag
e
i
n
t
he
are
as
of
l
o
w
c
o
l
or
v
ari
ati
on
s
th
at
woul
d i
m
prov
e s
e
gm
en
tat
i
o
n.
3.2
.
Co
n
t
o
u
r
Bas
ed F
ea
t
u
r
es
S
egm
ent
atio
n
Cont
o
ur
f
ea
tures
are
m
etri
c
s
uti
l
i
z
ed
t
o
ex
tr
ac
t
i
nf
orm
ati
on
ab
o
ut
tex
ture
,
c
ol
o
r
an
d
brig
ht
ne
s
s
.
T
hi
s
s
ec
ti
on
de
p
i
c
ts
the
brig
htn
es
s
,
c
ol
or,
an
d
tex
ture
f
ea
t
ure
an
d
ho
w
i
t
i
s
c
o
m
pu
ted
ef
f
i
c
i
en
t
l
y
.
3.2.1
.
T
ex
t
u
r
e featu
r
e
Com
pu
ti
ng
t
hi
s
es
te
em
i
s
ba
s
ed
o
n
a
s
i
m
pl
e
c
om
pa
r
i
s
on
of
tex
t
on
di
s
tr
i
bu
t
i
o
ns
on
ei
th
er
s
i
de
of
a
pi
x
e
l
i
n
r
e
s
pe
c
t
to
i
ts
o
v
er
w
h
el
m
i
ng
orie
nt
ati
on
.
W
e
c
an
c
ha
n
g
e
o
v
er
th
i
s
to
l
i
k
el
i
h
oo
d
l
i
k
e e
s
tee
m
ut
i
l
i
z
i
ng
th
e f
un
c
ti
o
n a
s
t
ak
es
af
t
er
]
)
(
e
x
p[
1
1
1
~
2
−
−
+
−
=
LR
t
e
x
t
u
r
e
X
P
(
3)
T
hi
s
es
tee
m
,
w
hi
c
h
ex
t
en
d
s
be
t
w
ee
n
0
a
nd
1,
tex
ture
es
tee
m
i
s
l
i
ttl
e
i
f
the
di
s
tr
i
bu
ti
on
s
on
th
e
t
w
o
s
i
de
s
are
al
tog
e
the
r
d
i
f
f
erent
an
d
h
ug
e
oth
erw
i
s
e
a
nd
2
LR
X
i
s
the
m
ax
i
m
al
l
i
k
el
i
ho
od
es
tee
m
.
G
en
eral
l
y
,
1
~
=
t
e
x
t
u
r
e
P
f
or
s
i
tua
t
e
d
e
ne
r
g
y
m
ax
i
m
a
i
n
tex
ture
an
d
0
~
=
t
e
x
t
u
r
e
P
i
s
f
or c
on
tou
r
s
.
t
e
x
t
u
r
e
P
~
i
s
d
ef
i
ne
d
to
be
0
at
p
i
x
el
s
whi
c
h
are n
ot
s
i
tua
t
ed
en
erg
y
m
ax
i
m
a.
3.2.2
.
Bri
g
h
t
n
e
ss
A
f
ew
ob
j
ec
ts
i
n
the
i
m
ag
e
c
an
be
bl
ac
k
or
whi
t
e.
T
he
y
are
no
t
s
al
i
en
t
i
n
c
ol
or
bu
t
r
ath
er
i
n
bri
gh
t
ne
s
s
.
It
r
etu
r
ns
th
e
i
nt
en
s
i
t
y
m
ea
s
ure
of
brig
htn
es
s
be
t
w
ee
n
a
p
i
x
el
a
nd
i
ts
ne
i
g
hb
or
throug
h t
h
e
w
h
ol
e i
m
ag
e a
nd
, b
r
i
gh
t
ne
s
s
i
s
0
f
or a c
on
s
tan
t
i
m
ag
e.
−
=
b
a
b
a
P
b
a
B
,
2
)
,
(
ˆ
(
4)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
E
ffe
c
t
i
v
e s
eg
me
nt
ati
on
of
s
c
l
era, i
r
i
s
an
d p
u
pi
l
i
n n
oi
s
y
ey
e i
ma
g
es
(
M
r
un
a
l
P
at
ha
k
)
2349
In
w
h
i
c
h,
B
ˆ
i
s
th
e
brig
htn
es
s
an
d
)
,
(
b
a
P
i
s
th
e p
i
x
el
at
l
oc
ati
on
)
,
(
b
a
.
3.2.3
.
Co
lor f
ea
t
u
r
e
Col
or
f
ea
t
ure
i
nd
i
c
ate
s
the
r
ate
of
oc
c
urr
en
c
e
of
ea
c
h
c
ol
or
i
n
de
x
es
i
n
an
i
m
ag
e
w
i
t
h
di
s
s
i
m
i
l
ar i
nt
en
s
i
t
i
es
.
Col
or
f
ea
ture v
ec
tor f
or a
gi
v
en
i
m
ag
e i
s
s
eg
m
en
ted
b
y
the
c
on
di
t
i
o
n (5)
:
=
=
M
j
i
z
M
z
1
1
ˆ
(
5)
w
he
r
e,
M
i
s
th
e
qu
an
t
i
t
y
of
pi
x
el
s
w
i
thi
n e
ac
h b
l
oc
k
,
i
z
i
s
th
e p
i
x
e
l
i
nte
ns
i
t
y
.
3.2.
4.
E
n
t
r
o
p
y
E
ntro
p
y
(
En
)
i
s
uti
l
i
z
e
d
to
de
s
c
r
i
be
the
tex
ture
of
of
i
r
i
s
an
d
no
n
-
i
r
i
s
l
i
k
e
s
c
l
era
a
nd
pu
p
i
l
th
e
i
np
ut
i
m
ag
e.
H
ere
the
e
ntrop
y
i
s
ev
al
ua
te
d
f
o
r
the
s
eg
m
en
ted
c
on
t
ou
r
b
as
ed
f
ea
t
ures
.
E
ntro
p
y
of
th
e
th
i
s
up
er p
i
x
el
i
y
E
i
s
c
om
pu
ted
b
y
th
e c
on
d
i
t
i
o
n (6)
as
:
−
=
−
=
−
=
1
0
1
0
2
)))
,
(
(
l
o
g
)(
,
(
m
i
m
j
i
y
j
i
P
j
i
P
E
(
6)
w
he
r
e
i
and
j
are
th
e
c
oe
ff
i
c
i
en
ts
of
c
o
-
oc
c
urr
en
c
e
m
atri
x
,
)
,
(
j
i
P
i
s
the
c
om
po
ne
n
t
i
n
the
c
o
-
oc
c
urr
en
c
e
m
atri
x
at
the
c
o
ordi
na
t
es
i
,
j
and
N
i
s
th
e
d
i
m
en
s
i
o
n
of
the
c
o
-
oc
c
urr
en
c
e m
atri
x
.
3.3
.
S
egm
ent
atio
n
of
S
cl
e
r
a,
I
r
is
and
P
u
p
il
Regio
n
s
u
sing
CNN Clu
steri
n
g
Con
v
ol
uti
on
a
l
n
eu
r
a
l
ne
t
w
ork
s
are
ge
ne
r
a
l
l
y
m
ad
e
b
y
a
s
et
o
ut
of
l
a
y
ers
tha
t
c
an
be
ga
th
ered
b
y
the
i
r
f
un
c
ti
on
al
i
ti
es
.
T
he
ex
tr
ac
ted
en
tr
op
y
f
ea
tur
es
s
et
i
y
y
y
y
i
y
E
E
E
E
E
f
,
.
.
.
.
.
,
,
)
(
3
2
1
=
are
the
i
np
ut
to
t
he
CNN
c
l
as
s
i
f
i
er
to
s
eg
m
en
t
the
i
r
i
s
,
s
c
l
era
an
d
p
up
i
l
s
eg
m
en
ts
.
T
he
s
tr
uc
ture
of
Con
v
ol
u
ti
o
n n
eu
r
a
l
ne
t
wor
k
c
l
as
s
i
f
i
er i
s
a
p
pe
are
d i
n
F
i
gu
r
e
2
.
F
i
gu
r
e
2
.
S
tr
uc
ture
of
c
on
v
ol
ut
i
o
na
l
ne
ural
ne
t
wor
k
T
he
c
on
c
l
ud
i
ng
ou
tpu
t
de
c
i
s
i
on
of
the
CN
N
de
p
en
ds
on
the
wei
gh
ts
an
d
b
i
as
es
.
Cons
eq
ue
nt
l
y
, th
es
e a
r
e
up
da
te
d
w
i
th
c
on
di
t
i
o
n (7)
a
n
d (8)
c
orr
es
po
nd
i
n
gl
y
f
or ea
c
h l
a
y
er
:
)
(
t
W
m
W
C
n
x
W
r
x
W
l
l
l
l
+
−
−
=
(
7)
)
(
t
B
m
B
C
n
x
B
l
l
l
+
−
=
(
8)
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
2
3
46
-
2
3
54
2350
w
he
r
e
W
,
B
,
l
,
,
x
,
n
,
m
,
t
,
C
de
n
ote
s
the
wei
gh
t,
bi
as
,
l
a
y
er
n
um
be
r
,
r
eg
ul
ar
i
z
ati
on
pa
r
am
ete
r
,
l
ea
r
n
i
ng
r
at
e,
t
ota
l
n
um
be
r
of
t
r
ai
ni
ng
s
a
m
pl
es
,
m
o
m
en
tu
m
,
up
da
ti
ng
s
tep
,
c
os
t
f
un
c
ti
on
r
es
p
ec
ti
v
el
y
.
T
he
CNN c
l
as
s
i
f
i
er c
on
s
i
s
ts
of
di
f
f
erent t
y
pe
s
of
l
a
y
e
r
s
as
:
a)
Con
v
ol
uti
on
a
l
l
a
y
er:
T
hi
s
l
a
y
er
pe
r
f
orm
s
the
c
on
v
o
l
ut
i
o
n
on
the
i
n
pu
t
da
ta
wi
th
the
k
ernel
us
i
ng
(
9)
:
−
=
−
=
1
0
N
n
n
k
n
k
h
x
y
(
9)
h
ere
x
,
h
,
N
,
y
de
n
ote
s
the
i
np
ut
f
ea
tures
,
f
i
l
ter,
nu
m
be
r
of
el
em
en
ts
i
n
x
,
o
utp
u
t
r
es
pe
c
ti
v
el
y
.
b)
P
oo
l
i
ng
l
a
y
er
: T
he
p
oo
l
i
n
g l
a
y
er r
e
du
c
es
th
e d
i
m
en
s
i
on
of
ou
tp
ut
n
eu
r
o
ns
.
c)
F
ul
l
y
c
on
ne
c
te
d
l
a
y
e
r
:
T
hi
s
l
a
y
er
c
on
ne
c
ts
ev
ery
ne
uron
f
r
om
the
m
ax
-
po
ol
ed
l
a
y
e
r
to
e
v
er
y
on
e
of
th
e
o
utp
u
t n
e
urons
.
T
he
ac
ti
v
a
ti
o
n f
un
c
ti
on
us
e
d i
n
th
i
s
w
ork
i
s
as
f
ol
l
o
w
s
:
S
of
tm
ax
:
T
hi
s
f
un
c
ti
on
c
om
pu
tes
t
he
pro
ba
bi
l
i
t
y
d
i
s
tr
i
b
uti
o
n o
f
th
e
k
ou
tpu
t c
l
as
s
es
:
=
k
x
x
i
i
i
e
e
p
1
(
10
)
he
r
e
,
t
he
C
NN
c
l
us
ters
the
d
ata
i
n
i
m
ag
es
ba
s
e
d
on
the
en
tr
o
p
y
v
al
ue
an
d
r
es
u
l
ts
the
ef
f
ec
ti
v
e
s
c
l
era,
i
r
i
s
an
d
pu
p
i
l
r
e
gi
on
s
e
gm
en
tat
i
o
ns
.
4.
Re
sult
s a
n
d
A
n
al
y
s
is
T
he
propos
e
d
ef
f
i
c
i
en
t
s
eg
m
en
tat
i
on
of
i
r
i
s
,
s
c
l
era
an
d
p
up
i
l
r
e
gi
on
s
b
y
ut
i
l
i
z
i
n
g
en
tr
op
y
ba
s
ed
CNN
c
l
us
teri
n
g
i
s
i
m
pl
e
m
en
ted
i
n
M
A
T
LA
B
.
T
he
fr
ee
l
y
ac
c
es
s
i
b
l
e
M
M
U
da
tab
as
e
an
d
UB
IRI
S
.
v
2
da
t
ab
as
e
of
e
y
e
i
m
ag
es
are
ut
i
l
i
z
ed
to
as
s
es
s
propos
ed
s
eg
m
en
tat
i
o
ns
.
In
th
i
s
s
ec
ti
on
,
the
ex
pe
r
i
m
en
ta
l
r
es
ul
ts
ac
c
om
pl
i
s
he
d
f
or the
propo
s
e
d
tec
h
ni
qu
e
are g
i
v
en
.
F
i
gu
r
e
3
an
d
F
i
gu
r
e
4
de
pi
c
ts
the
s
eg
m
en
tat
i
o
n
of
i
r
i
s
,
s
c
l
era
an
d
pu
p
i
l
f
or
the
n
um
be
r
of
i
np
u
t
s
am
pl
e
e
y
e
i
m
ag
es
t
ak
en
f
r
o
m
the
M
MU
d
ata
ba
s
e
an
d
UB
IRI
S
.v
2
d
ata
ba
s
e
r
es
pe
c
ti
v
el
y
.
T
he
c
om
pa
r
i
s
on
T
ab
l
e
1,
T
ab
l
e
2,
an
d
T
ab
l
e
3
i
l
l
us
tr
ate
s
tha
t
o
ur
prop
os
ed
i
r
i
s
,
s
c
l
era
an
d
pu
pi
l
s
eg
m
en
tat
i
on
uti
l
i
z
i
ng
CNN
i
s
ex
t
e
ns
i
v
e
l
y
be
tt
er
th
an
t
he
ex
i
s
ti
n
g
A
NF
I
S
an
d
K
NN.
T
he
c
om
pa
r
i
s
on
graph
of
s
eg
m
en
tat
i
o
ns
r
eg
ards
of
ac
c
urac
y
,
s
en
s
i
ti
v
i
t
y
,
P
P
V
a
nd
s
pe
c
i
f
i
c
i
t
y
(
F
NR
an
d
F
DR)
of
i
r
i
s
,
s
c
l
era
an
d
p
up
i
l
are
s
ho
w
n
i
n
F
i
g
ur
es
5
-
6,
7
-
8,
9
-
10
r
es
pe
c
t
i
v
el
y
.
(
a)
(
b)
(
c
)
(
d)
(
e)
(
f
)
F
i
gu
r
e
3.
S
eg
m
en
tat
i
on
of
(
a) Inpu
t
i
m
ag
e,
(
b) c
ou
nte
r
i
m
ag
e,
(
c
)
en
tr
op
y
i
m
ag
e,
(
d)
pu
p
i
l
,
(
e)
i
r
i
s
, a
n
d (f
)
s
c
l
era r
eg
i
on
s
i
m
ag
es
ta
k
en
f
r
o
m
MM
U
da
ta
ba
s
e
(
a)
(
b)
(
c
)
(
d)
(
e)
(
f
)
F
i
gu
r
e
4.
S
eg
m
en
tat
i
on
of
(
a) i
np
ut
i
m
ag
e,
(
b) c
ou
nt
er i
m
ag
e,
(
c
)
en
tr
op
y
i
m
ag
e,
(
d) pup
i
l
,
(
e) i
r
i
s
, a
n
d (f
)
s
c
l
era r
eg
i
on
s
i
m
ag
es
ta
k
en
f
r
o
m
UB
IRI
S
.v
2 d
ata
b
as
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
E
ffe
c
t
i
v
e s
eg
me
nt
ati
on
of
s
c
l
era, i
r
i
s
an
d p
u
pi
l
i
n n
oi
s
y
ey
e i
ma
g
es
(
M
r
un
a
l
P
at
ha
k
)
2351
4.1
.
I
r
is
S
egm
ent
atio
n
T
he
c
o
m
pa
r
i
s
on
ta
bl
e
of
p
r
op
os
ed
i
r
i
s
s
e
gm
en
tat
i
o
n
wi
th
ex
i
s
ti
ng
A
NF
I
S
,
an
d
K
NN
i
n
r
eg
ards
to
di
f
f
erent e
x
ec
ut
i
on
m
ea
s
ures
i
s
po
r
tr
a
y
ed
i
n
T
ab
l
e 1
.
T
ab
l
e 1
. C
om
pa
r
i
s
on
A
na
l
y
s
i
s
of
P
r
op
os
ed
Me
tho
d
i
n
T
er
m
s
of
V
ario
us
P
erf
or
m
a
nc
e M
e
as
ures
D
a
t
a
b
a
s
e
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
S
e
n
s
i
t
iv
it
y
S
p
e
c
i
f
i
c
it
y
P
P
V
FN
R
FD
R
MMU
P
r
o
p
o
s
e
d
9
7
.
1
4
5
9
9
.
6
9
4
1
1
.
1
3
0
9
7
.
4
2
6
0
.
3
0
5
2
.
5
7
3
A
N
FI
S
9
6
.
3
0
6
9
9
.
7
6
1
1
0
.
0
8
1
9
6
.
5
1
4
0
.
2
3
8
3
.
4
8
5
K
N
N
9
5
.
9
3
1
9
9
.
7
9
8
9
.
8
8
0
9
6
.
1
0
0
0
.
2
0
1
3
.
8
9
9
U
B
I
R
I
S
.
v
2
P
r
o
p
o
s
e
d
9
7
.
9
9
6
9
8
.
1
0
1
5
4
.
1
7
1
9
9
.
8
3
7
1
.
8
9
8
0
.
1
6
2
A
N
FI
S
9
3
.
2
0
3
9
7
.
2
9
5
2
9
.
9
7
7
9
5
.
5
5
0
2
.
7
0
4
4
.
4
4
9
K
N
N
9
5
.
6
3
4
9
5
.
6
4
0
3
3
.
3
3
3
9
6
.
9
9
3
4
.
3
5
9
1
.
0
0
6
(
a)
(
b)
F
i
gu
r
e
5.
C
om
pa
r
i
s
on
grap
h o
f
i
r
i
s
s
eg
m
en
tat
i
on
i
n t
er
m
s
of
ac
c
urac
y
, s
en
s
i
t
i
v
i
t
y
an
d
P
P
V
f
or (
a) MMU d
at
ab
as
e (b)
U
B
IRI
S
.v
2 d
ata
ba
s
e
(
a)
(
b)
F
i
gu
r
e
6.
C
om
pa
r
i
s
on
grap
h f
or i
r
i
s
s
eg
m
en
tat
i
on
i
n t
e
r
m
s
of
s
pe
c
i
f
i
c
i
t
y
, FNR
an
d
F
DR
f
or (
a) MMU
da
t
ab
as
e (b)
U
B
IRI
S
.v
2 d
ata
ba
s
e
4.2
.
S
cl
er
a
S
egm
ent
atio
n
T
he
c
om
pa
r
i
s
on
tab
l
e
of
pr
op
os
ed
s
c
l
era
s
eg
m
en
tat
i
o
n
wi
th
ex
i
s
t
i
ng
A
NF
IS
,
a
nd
K
N
N
i
n reg
ards
to
di
f
f
erent e
x
ec
uti
o
n m
ea
s
ures
i
s
po
r
tr
a
y
e
d i
n
T
ab
l
e 2
.
T
ab
l
e
2
.
C
om
pa
r
i
s
on
A
na
l
y
s
i
s
of
P
r
op
os
ed
T
ec
hn
i
qu
e
i
n
T
er
m
s
of
Di
ff
erent P
erf
or
m
an
c
e M
ea
s
ures
D
a
t
a
b
a
s
e
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
S
e
n
s
i
t
iv
it
y
S
p
e
c
i
f
i
c
it
y
P
P
V
FN
R
FD
R
MMU
P
r
o
p
o
s
e
d
9
5
.
5
3
8
9
9
.
2
1
0
7
.
5
0
4
9
6
.
2
5
7
0
.
7
8
9
3
.
7
4
2
A
N
FI
S
9
6
.
6
4
0
9
9
.
1
2
5
7
.
6
5
2
9
7
.
4
6
4
0
.
8
7
4
2
.
5
3
5
K
N
N
9
5
.
3
4
4
9
9
.
1
8
3
6
.
6
5
2
9
6
.
0
8
5
0
.
8
1
6
3
.
9
1
4
U
B
I
R
I
S
.
v
2
P
r
o
p
o
s
e
d
9
8
.
0
8
0
9
9
.
6
6
3
1
8
.
5
7
1
9
8
.
3
7
7
0
.
3
3
6
1
.
6
2
2
A
N
FI
S
8
8
.
0
0
2
9
6
.
5
8
8
5
.
1
1
9
8
9
.
0
5
2
1
.
4
1
1
1
0
.
9
4
7
K
N
N
8
2
.
3
6
3
9
8
.
4
8
6
7
.
5
8
3
8
2
.
4
6
0
0
.
5
1
3
1
7
.
5
3
9
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
2
3
46
-
2
3
54
2352
(
a)
(
b)
F
i
gu
r
e
7
.
C
om
pa
r
i
s
on
grap
h f
or s
c
l
era s
eg
m
en
tat
i
o
n i
n t
erm
s
of
ac
c
urac
y
, s
e
ns
i
t
i
v
i
t
y
an
d
P
P
V
f
or (
a) MMU d
at
ab
as
e (b)
U
B
IRI
S
.v
2 d
ata
ba
s
e
(
a)
(
b)
F
i
gu
r
e
8.
C
om
pa
r
i
s
on
grap
h f
or s
c
l
era s
eg
m
en
tat
i
o
n i
n t
erm
s
of
s
pe
c
i
f
i
c
i
t
y
, FNR
an
d FDR
f
or (
a) MMU d
at
ab
as
e (b)
U
B
IRI
S
.v
2 d
ata
ba
s
e
4.3
.
P
u
p
il
S
egm
ent
atio
n
T
he
c
om
pa
r
i
s
on
tab
l
e
of
pr
op
os
ed
p
up
i
l
s
e
gm
en
tat
i
o
n
w
i
t
h
ex
i
s
ti
n
g
A
NF
I
S
,
an
d
K
NN
i
n
r
eg
ards
to
of
di
f
f
erent
ex
ec
uti
o
n m
ea
s
ures
i
s
po
r
tr
a
y
e
d i
n
T
ab
l
e 3
.
T
ab
l
e 3
. C
om
pa
r
i
s
on
A
na
l
y
s
i
s
of
P
r
op
os
ed
Me
tho
d
i
n
T
e
r
m
s
of
Di
ff
erent P
erf
or
m
an
c
e M
ea
s
ures
D
a
t
a
b
a
s
e
M
e
t
h
o
d
s
A
c
c
u
r
a
c
y
S
e
n
s
i
t
iv
it
y
S
p
e
c
i
f
i
c
it
y
P
P
V
FN
R
FD
R
MMU
P
r
o
p
o
s
e
d
9
8
.
2
7
5
9
9
.
9
5
7
7
.
3
6
3
9
8
.
3
1
4
0
.
0
4
2
1
.
6
8
5
A
N
FI
S
9
5
.
9
4
2
9
9
.
9
7
4
3
.
6
0
5
9
5
.
9
6
0
0
.
0
2
5
9
4
.
0
3
9
K
N
N
9
7
.
9
4
7
9
9
.
9
6
8
6
.
6
7
6
9
7
.
9
7
4
0
.
0
3
1
7
2
.
0
2
5
U
B
I
R
I
S
.
v
2
P
r
o
p
o
s
e
d
9
9
.
4
2
3
9
9
.
4
9
5
3
3
.
9
1
3
9
9
.
9
2
6
0
.
5
0
4
0
.
0
7
3
A
N
FI
S
9
7
.
8
2
1
9
6
.
7
2
8
1
8
.
8
8
7
9
8
.
0
7
2
0
.
8
7
1
1
.
9
2
7
K
N
N
9
1
.
3
8
9
9
7
.
8
9
9
6
.
7
7
9
9
1
.
4
2
0
0
.
7
0
0
8
.
5
7
9
(
a)
(
b)
F
i
gu
r
e
9.
C
om
pa
r
i
s
on
grap
h f
or pup
i
l
s
e
gm
en
tat
i
o
n i
n
term
s
of
ac
c
urac
y
, s
en
s
i
ti
v
i
t
y
a
nd
P
P
V
f
or
(
a) MMU
da
ta
ba
s
e (b)
U
B
I
RIS
.
v
2
da
ta
ba
s
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
E
ffe
c
t
i
v
e s
eg
me
nt
ati
on
of
s
c
l
era, i
r
i
s
an
d p
u
pi
l
i
n n
oi
s
y
ey
e i
ma
g
es
(
M
r
un
a
l
P
at
ha
k
)
2353
(
a)
(
b)
F
i
gu
r
e
10
.
Com
pa
r
i
s
on
gra
ph
f
or pup
i
l
s
eg
m
en
tat
i
on
i
n t
erm
s
of
s
pe
c
i
f
i
c
i
t
y
, FNR
an
d FDR
f
or
(
a) MMU
da
ta
ba
s
e (b)
U
B
I
RIS
.
v
2
da
ta
ba
s
e
5
. Con
clus
ion
In
t
hi
s
pa
p
er
we
ha
v
e
pr
es
en
te
d
a
pr
of
i
c
i
en
t
s
eg
m
en
tat
i
o
n
of
i
r
i
s
,
s
c
l
era
an
d
p
up
i
l
r
eg
i
o
ns
f
r
om
the
e
y
e
i
m
ag
es
i
nd
i
v
i
du
al
l
y
u
ti
l
i
z
i
ng
en
tr
o
p
y
b
as
ed
C
NN
c
l
us
t
erin
g.
T
he
pe
r
f
or
m
an
c
e
of
i
r
i
s
,
s
c
l
era
an
d
s
eg
m
en
tat
i
on
i
s
hi
gh
l
y
d
ep
e
nd
s
o
n
th
e
s
eg
m
en
tat
i
o
n
proc
es
s
.
Her
e,
CNN
ef
f
e
c
ti
v
e
l
y
s
eg
m
en
ts
the
i
r
i
s
,
s
c
l
era
r
eg
i
on
s
ba
s
e
d
o
n
the
en
tr
op
y
m
ea
s
ures
.
T
he
ex
pe
r
i
m
en
tal
o
u
tc
om
es
ex
hi
bi
t
t
ha
t
ou
r
prop
os
ed
c
l
as
s
i
f
i
c
at
i
on
ou
t
pe
r
f
orm
s
the
ex
i
s
ti
ng
K
NN
an
d
A
NF
I
S
m
eth
od
s
i
n
r
eg
ards
of
pe
r
f
or
m
an
c
e
m
ea
s
ures
s
uc
h
as
,
ac
c
urac
y
,
s
en
s
i
ti
v
i
t
y
,
s
pe
c
i
f
i
c
i
t
y
,
P
P
V
,
NP
V
, FP
R, FN
R, FD
R, F
-
m
ea
s
ure an
d M
CC
.
Ref
er
en
ce
s
[1
]
C
W
T
a
n
,
A
Ku
m
a
r.
Au
t
o
m
a
te
d
s
e
g
m
e
n
ta
ti
o
n
o
f
i
ri
s
i
m
a
g
e
s
u
s
i
n
g
v
i
s
i
b
l
e
wa
v
e
l
e
n
g
t
h
fa
c
e
i
m
a
g
e
s
.
IEEE
Com
p
u
te
r So
c
i
e
ty
Co
n
fe
re
n
c
e
o
n
.
IEEE
.
2
0
1
1
:
9
-
14
.
[2
]
CW
T
a
n
,
A
Ku
m
a
r
.
Uni
f
i
e
d
fra
m
e
w
o
rk
fo
r
a
u
t
o
m
a
te
d
i
ri
s
s
e
g
m
e
n
ta
ti
o
n
u
s
i
n
g
d
i
s
ta
n
t
l
y
a
c
q
u
i
re
d
fa
c
e
i
m
a
g
e
s
.
IEEE
Tra
n
s
a
c
ti
o
n
s
o
n
I
m
a
g
e
Pro
c
e
s
s
i
n
g
.
2
0
1
2
.
[3
]
M
K
Pa
th
a
k
,
N
Sri
n
i
v
a
s
u
,
VK
Ba
i
ra
g
i
.
M
a
tc
h
s
c
o
re
l
e
v
e
l
fu
s
i
o
n
o
f
i
ri
s
a
n
d
s
c
l
e
r
a
d
e
s
c
r
i
p
to
r
fo
r
i
ri
s
re
c
o
g
n
i
ti
o
n
.
IEEE
In
te
rn
a
t
i
o
n
a
l
Co
n
fe
re
n
c
e
o
n
S
o
ft
Com
p
u
ti
n
g
a
n
d
i
t
s
E
n
g
i
n
e
e
r
i
n
g
A
p
p
l
i
c
a
t
i
o
n
s
.
2017
:
1
-
6.
[4
]
RR
J
i
l
l
e
l
a
,
A
Ro
s
s
.
Se
g
m
e
n
ti
n
g
i
ri
s
i
m
a
g
e
s
i
n
th
e
v
i
s
i
b
l
e
s
p
e
c
tr
u
m
w
i
th
a
p
p
l
i
c
a
ti
o
n
s
i
n
m
o
b
i
l
e
b
i
o
m
e
tri
c
s
.
Pa
tt
e
r
n
Re
c
o
g
n
i
ti
o
n
L
e
tt
e
rs
.
2
0
1
5
;
5
7
:
4
-
1
6
.
[5
]
C
W
T
a
n
,
A
Ku
m
a
r.
T
o
w
a
rd
s
o
n
l
i
n
e
i
ri
s
a
n
d
p
e
r
i
o
c
u
l
a
r
re
c
o
g
n
i
t
i
o
n
u
n
d
e
r
re
l
a
x
e
d
i
m
a
g
i
n
g
c
o
n
s
tra
i
n
ts
.
IEEE
Tra
n
s
a
c
ti
o
n
s
o
n
Im
a
g
e
Pro
c
e
s
s
i
n
g
.
2
0
1
3
;
22
(1
0
):
3
7
5
1
-
3
7
6
5
.
[6
]
M
M
a
h
l
o
u
j
i
,
A
Noru
z
i
.
Hum
a
n
i
r
i
s
s
e
g
m
e
n
t
a
ti
o
n
f
o
r
i
ri
s
re
c
o
g
n
i
ti
o
n
i
n
u
n
c
o
n
s
tr
a
i
n
e
d
e
n
v
i
ro
n
m
e
n
t
s
.
I
n
te
rn
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Com
p
u
t
e
r Sc
i
e
n
c
e
I
s
s
u
e
s
(I
J
CSI)
.
2
0
1
2
;
9
(
1
):
1
4
9
.
[7
]
C
Rat
h
g
e
b
,
A
Uhl
,
P
W
i
l
d
.
Iri
s
b
i
o
m
e
tri
c
s
:
fro
m
s
e
g
m
e
n
t
a
ti
o
n
to
t
e
m
p
l
a
t
e
s
e
c
u
ri
ty
.
Sp
ri
n
g
e
r
Sc
i
e
n
c
e
&
Bu
s
i
n
e
s
s
M
e
d
i
a
.
2
0
1
2
:
5
9
.
[8
]
DS
J
e
o
n
g
,
JW
Hw
a
n
g
,
BJ
Ka
n
g
,
KR
Pa
rk
,
CS
W
o
n
,
DK
Pa
r
k
,
J
Ki
m
.
A
n
e
w
i
ri
s
s
e
g
m
e
n
t
a
ti
o
n
m
e
th
o
d
fo
r n
o
n
-
i
d
e
a
l
i
ri
s
i
m
a
g
e
s
.
Im
a
g
e
a
n
d
v
i
s
i
o
n
c
o
m
p
u
ti
n
g
.
2
0
1
0
;
28
(2
)
:
254
-
2
6
0
.
[9
]
MA
Ab
d
u
l
l
a
h
,
SS
Dl
a
y
,
WL
W
o
o
,
JA
Cha
m
b
e
r
s
.
Rob
u
s
t
i
ri
s
s
e
g
m
e
n
ta
ti
o
n
m
e
th
o
d
b
a
s
e
d
o
n
a
n
e
w
a
c
ti
v
e
c
o
n
t
o
u
r f
o
rc
e
wit
h
a
n
o
n
c
i
rc
u
l
a
r
n
o
rm
a
l
i
z
a
ti
o
n
.
IEEE
S
M
C
.
2
0
1
6
.
[1
0
]
Y
Che
n
,
M
Ad
j
o
u
a
d
i
,
C
Ha
n
,
J
W
a
n
g
,
A
Ba
rre
t
o
,
N
Ris
h
e
,
J
An
d
ri
a
n
.
A
h
i
g
h
l
y
a
c
c
u
ra
t
e
a
n
d
c
o
m
p
u
ta
t
i
o
n
a
l
l
y
e
f
fi
c
i
e
n
t
a
p
p
ro
a
c
h
f
o
r
u
n
c
o
n
s
tra
i
n
e
d
i
ri
s
s
e
g
m
e
n
t
a
ti
o
n
.
Im
a
g
e
a
n
d
Vi
s
i
o
n
Com
p
u
ti
n
g
.
2
0
1
0
;
28
(2
)
:
2
6
1
-
2
6
9
.
[1
1
]
D
O
Roi
g
,
AM
G
o
n
z
á
l
e
z
,
EG
Llano
.
Se
m
a
n
t
i
c
s
e
g
m
e
n
ta
t
i
o
n
o
f
c
o
l
o
r
e
y
e
i
m
a
g
e
s
fo
r
i
m
p
ro
v
i
n
g
i
ri
s
s
e
g
m
e
n
t
a
ti
o
n
.
Ib
e
ro
a
m
e
ri
c
a
n
Con
g
re
s
s
o
n
Pa
tt
e
rn
Re
c
o
g
n
i
t
i
o
n
.
Sp
r
i
n
g
e
r
,
Cha
m
.
2
0
1
7
:
4
6
6
-
474
.
[1
2
]
R
M
e
s
b
a
h
,
B
M
c
Can
e
,
S
M
i
l
l
s
.
Co
n
d
i
t
i
o
n
a
l
ra
n
d
o
m
fi
e
l
d
s
i
n
c
o
rp
o
ra
te
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
r
a
l
n
e
two
r
k
s
fo
r
h
u
m
a
n
e
y
e
s
c
l
e
r
a
s
e
m
a
n
ti
c
s
e
g
m
e
n
ta
ti
o
n
.
B
i
o
m
e
tri
c
s
(I
J
CB),
In
te
rn
a
t
i
o
n
a
l
J
o
i
n
t
Co
n
fe
re
n
c
e
o
n
.
IEEE
.
2
0
1
7
.
[1
3
]
C
Fa
ra
b
e
t,
C
Co
u
p
ri
e
,
L
Na
j
m
a
n
,
Y
L
e
Cun
.
L
e
a
rn
i
n
g
h
i
e
r
a
rc
h
i
c
a
l
f
e
a
tu
r
e
s
fo
r
s
c
e
n
e
l
a
b
e
l
l
i
n
g
.
IEE
E
Tra
n
s
.
Pa
tt
e
r
n
An
a
l
.
M
a
c
h
.
In
t
e
l
l
.
2
0
1
3
;
3
5
(
8
)
: 1
-
15
.
[1
4
]
Z
Zh
o
u
,
E
Y
Du,
NL
T
h
o
m
a
s
,
EJ
Del
p
.
A
New
Hum
a
n
Id
e
n
ti
f
i
c
a
t
i
o
n
M
e
th
o
d
:
S
c
l
e
r
a
Rec
o
g
n
i
ti
o
n
.
IEE
E
Tra
n
s
a
c
t
i
o
n
s
o
n
Sy
s
te
m
s
,
M
a
n
,
a
n
d
Cy
b
e
rn
e
ti
c
s
-
Pa
rt
A:
Sy
s
te
m
s
a
n
d
Hum
a
n
s
.
2
0
1
2
;
42
(
3
):
571
-
5
8
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
2
3
46
-
2
3
54
2354
[1
5
]
Raj
o
l
e
,
Sn
e
h
a
l
S
,
J
V
Sh
i
n
d
e
.
In
te
g
ra
te
d
A
p
p
ro
a
c
h
fo
r
S
c
l
e
r
a
Rec
o
g
n
i
t
i
o
n
a
n
d
Ey
e
G
a
z
e
Det
e
c
t
i
o
n
.
2
0
1
7
;
6
(7
):
7
7
-
82
.
[1
6
]
A
Das
,
U
Pa
l
,
M
A
Fe
rre
rBa
l
l
e
s
te
r
,
M
Bl
u
m
e
n
s
t
e
i
n
.
A
n
e
w
e
ff
i
c
i
e
n
t
a
n
d
a
d
a
p
ti
v
e
s
c
l
e
ra
re
c
o
g
n
i
ti
o
n
s
y
s
te
m
.
Co
m
p
u
t
a
ti
o
n
a
l
In
t
e
l
l
i
g
e
n
c
e
i
n
Bi
o
m
e
tri
c
s
a
n
d
Id
e
n
ti
ty
M
a
n
a
g
e
m
e
n
t
,
IEEE
Sy
m
p
o
s
i
u
m
o
n
.
2014.
[1
7
]
P
Rad
u
,
J
F
e
rry
m
a
n
,
P
W
i
l
d
.
A
ro
b
u
s
t
s
c
l
e
ra
s
e
g
m
e
n
t
a
ti
o
n
a
l
g
o
r
i
th
m
Bi
o
m
e
tr
i
c
s
Th
e
o
ry
.
Ap
p
l
i
c
a
ti
o
n
s
a
n
d
Sy
s
te
m
s
(BT
AS),
IEEE
Con
fe
re
n
c
e
.
2
0
1
5
.
[1
8
]
SA
Sa
h
m
o
u
d
,
I
S
Ab
u
h
a
i
b
a
.
Eff
i
c
i
e
n
t
i
ri
s
s
e
g
m
e
n
ta
ti
o
n
m
e
th
o
d
i
n
u
n
c
o
n
s
tra
i
n
e
d
e
n
v
i
ro
n
m
e
n
ts
.
Pa
tt
e
rn
Re
c
o
g
n
i
t
i
o
n
.
2
0
1
3
;
4
6
(
1
)
:
3
1
7
4
-
3
1
8
5
.
[1
9
]
G
Sa
n
to
s
,
E
H
o
y
l
e
.
A
fu
s
i
o
n
a
p
p
ro
a
c
h
to
u
n
c
o
n
s
tra
i
n
e
d
i
ri
s
re
c
o
g
n
i
ti
o
n
.
P
a
tt
e
rn
R
e
c
o
g
n
i
t
i
o
n
L
e
t
te
rs
.
2
0
1
2
;
3
3
(1
)
:
9
8
4
-
9
9
0
.
[2
0
]
R
Sz
e
w
c
z
y
k
,
K
G
g
ra
b
o
w
s
k
i
,
M
Nap
i
e
ra
l
s
k
a
,
W
Sa
n
k
o
w
s
k
i
,
Zu
b
e
rt,
A
Nap
i
e
ra
l
s
k
i
.
A
r
e
l
i
a
b
l
e
i
r
i
s
re
c
o
g
n
i
ti
o
n
a
l
g
o
ri
t
h
m
b
a
s
e
d
o
n
re
v
e
rs
e
b
i
o
rt
h
o
g
o
n
a
l
w
a
v
e
l
e
t
tra
n
s
fo
r
m
.
Pa
tt
e
rn
R
e
c
o
g
n
i
t
i
o
n
L
a
ty
t
e
r
s
.
2012
;
3
3
(1
)
:
1
0
1
9
-
1
0
2
6
.
[2
1
]
Kri
s
h
n
a
,
N
M
u
ra
l
i
,
PCS
Red
d
y
.
A
Dim
e
n
s
i
o
n
a
l
i
ty
Red
u
c
e
d
Iri
s
Re
c
o
g
n
i
ti
o
n
Sy
s
te
m
w
i
th
Ai
d
o
f
A
I
T
e
c
h
n
i
q
u
e
s
.
G
l
o
b
a
l
J
o
u
rn
a
l
o
f
Res
e
a
r
c
h
i
n
En
g
i
n
e
e
ri
n
g
.
2
0
1
4
;
1
4
(4
):
1
1
-
25
.
[2
2
]
AF
Ab
a
te
,
e
t
a
l
.
Ku
rt
o
s
i
s
a
n
d
s
k
e
w
n
e
s
s
a
t
p
i
x
e
l
l
e
v
e
l
a
s
i
n
p
u
t
fo
r
SO
M
n
e
tw
o
rk
s
to
i
r
i
s
re
c
o
g
n
i
t
i
o
n
o
n
m
o
b
i
l
e
d
e
v
i
c
e
s
.
Pa
tt
e
rn
Re
c
o
g
n
i
ti
o
n
L
e
tt
e
r
s
.
2
0
1
7
;
91
:
37
-
4
3
.
[2
3
]
NF
So
l
i
m
a
n
,
E
M
o
h
a
m
e
d
,
F
M
a
g
d
i
,
F
E
Ab
d
El
-
S
a
m
i
e
,
M
Ab
d
El
n
a
b
y
.
Eff
i
c
i
e
n
t
i
ri
s
l
o
c
a
l
i
z
a
ti
o
n
a
n
d
re
c
o
g
n
i
ti
o
n
.
O
p
t
i
k
-
In
te
r
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
f
o
r L
i
g
h
t
a
n
d
E
l
e
c
tro
n
O
p
ti
c
s
.
2
0
1
7
:
469
-
4
7
5
.
[2
4
]
Ab
d
u
l
l
a
h
,
M
o
h
a
m
m
e
d
AM
,
S
S
Dla
y
,
W
L
W
o
o
,
J
A.
Ch
a
m
b
e
rs
.
Rob
u
s
t
i
ri
s
s
e
g
m
e
n
ta
ti
o
n
m
e
th
o
d
b
a
s
e
d
o
n
a
n
e
w
a
c
ti
v
e
c
o
n
to
u
r
f
o
rc
e
w
i
th
a
n
o
n
c
i
rc
u
l
a
r
n
o
rm
a
l
i
z
a
t
i
o
n
.
IEEE
tra
n
s
a
c
ti
o
n
s
o
n
s
y
s
te
m
s
.
2
0
1
7
;
47
(
12
)
:
3
1
2
8
-
3
1
4
1
.
[2
5
]
S
Al
k
a
s
s
a
r,
e
t
a
l
.
Ro
b
u
s
t
s
c
l
e
ra
r
e
c
o
g
n
i
t
i
o
n
s
y
s
te
m
w
i
th
n
o
v
e
l
s
c
l
e
ra
s
e
g
m
e
n
ta
t
i
o
n
a
n
d
v
a
l
i
d
a
ti
o
n
te
c
h
n
i
q
u
e
s
.
IEEE
Tra
n
s
a
c
ti
o
n
s
o
n
S
y
s
te
m
s
.
2
0
1
7
;
4
7
(3
):
474
-
4
8
6
.
[2
6
]
PR
Nal
l
a
,
A
K
u
m
a
r
.
T
o
w
a
rd
m
o
r
e
a
c
c
u
ra
te
i
r
i
s
r
e
c
o
g
n
i
ti
o
n
u
s
i
n
g
c
ro
s
s
-
s
p
e
c
tra
l
m
a
tc
h
i
n
g
.
IEE
E
tra
n
s
a
c
t
i
o
n
s
o
n
Im
a
g
e
p
r
o
c
e
s
s
i
n
g
.
2
0
1
7
;
26
(
1
):
2
0
8
-
2
2
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.