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.
1
,
F
e
br
ua
r
y
2020
,
pp.
427
~
435
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
.
v18i1.
12992
427
Jou
r
n
al
h
omepage
:
ht
tp:
//
jour
nal.
uad
.
ac
.
id/
index
.
php/T
E
L
K
OM
N
I
K
A
N
e
u
r
o
-
f
u
z
z
y i
n
f
e
r
e
n
c
e
syste
m
b
ase
d
f
ac
e
r
e
c
og
n
ition
u
si
n
g f
e
at
u
r
e
e
xt
r
ac
t
io
n
Ham
s
a
A.
Ab
d
u
ll
ah
Co
l
l
eg
e
o
f
I
n
fo
rma
t
i
o
n
E
n
g
i
n
eer
i
n
g
,
A
l
-
N
ah
r
ai
n
U
n
i
v
er
s
i
t
y
,
Iraq
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
Apr
23
,
2019
R
e
vis
e
d
J
un
3
0
,
20
19
Ac
c
e
pte
d
J
ul
18
,
20
19
H
u
ma
n
face
reco
g
n
i
t
i
o
n
(H
FR)
i
s
t
h
e
met
h
o
d
o
f
reco
g
n
i
zi
n
g
p
eo
p
l
e
i
n
i
ma
g
es
o
r
v
i
d
e
o
s
.
T
h
ere
are
d
i
fferen
t
H
FR
met
h
o
d
s
s
u
c
h
as
fea
t
u
re
-
b
as
ed
,
ei
g
e
n
-
faces
,
h
i
d
d
e
n
mark
o
v
m
o
d
el
a
n
d
n
eu
ra
l
n
et
w
o
r
k
(N
N
)
b
as
e
d
met
h
o
d
s
.
Feat
u
re
ex
t
rac
t
i
o
n
o
r
p
re
p
ro
ce
s
s
i
n
g
u
s
ed
i
n
fi
rs
t
t
h
ree
men
t
i
o
n
ed
met
h
o
d
s
t
h
a
t
as
s
o
ci
a
t
ed
w
i
t
h
t
h
e
cat
eg
o
ry
o
f
t
h
e
i
mag
e
t
o
reco
g
n
i
ze.
W
h
i
l
e
i
n
t
h
e
N
N
met
h
o
d
,
an
y
t
y
p
e
o
f
i
mag
e
ca
n
b
e
u
s
efu
l
w
i
t
h
o
u
t
t
h
e
req
u
i
reme
n
t
t
o
p
art
i
cu
l
ar
d
at
a
ab
o
u
t
t
h
e
t
y
p
e
o
f
i
mag
e,
an
d
s
i
m
u
l
t
an
e
o
u
s
l
y
p
ro
v
i
d
e
s
s
u
p
eri
o
r
ac
cu
racy
.
In
t
h
i
s
p
ap
er,
H
FR
s
y
s
t
em
b
a
s
ed
o
n
n
e
u
ral
-
f
u
zzy
(N
F)
h
as
b
ee
n
i
n
t
ro
d
u
ce
d
.
In
t
h
e
N
N
s
y
s
t
em,
b
ac
k
p
r
o
p
a
g
at
i
o
n
(BP)
al
g
o
ri
t
h
m
i
s
u
s
e
d
t
o
u
p
d
at
e
t
h
e
w
ei
g
h
t
s
o
f
t
h
e
n
eu
r
o
n
s
t
h
r
o
u
g
h
s
u
p
er
v
i
s
ed
l
earn
i
n
g
.
T
w
o
s
et
s
o
f
t
h
e
i
mag
e
h
av
e
b
een
u
s
ed
f
o
r
t
ra
i
n
i
n
g
an
d
t
es
t
i
n
g
t
h
e
n
et
w
o
r
k
t
o
i
d
en
t
i
f
y
t
h
e
p
ers
o
n
.
If
t
h
e
t
es
t
i
ma
g
e
mat
c
h
es
t
o
o
n
e
o
f
t
h
e
t
ra
i
n
ed
s
e
t
s
o
f
t
h
e
i
ma
g
e,
t
h
e
n
t
h
e
s
y
s
t
em
w
i
l
l
ret
u
r
n
reco
g
n
i
zed
.
A
n
d
i
f
t
h
e
t
e
s
t
i
mag
e
d
o
e
s
n
o
t
mat
c
h
t
o
o
n
e
o
f
t
h
e
t
ra
i
n
e
d
s
e
t
s
o
f
t
h
e
i
ma
g
e,
t
h
e
n
t
h
e
s
y
s
t
em
w
i
l
l
re
t
u
r
n
n
o
t
reco
g
n
i
ze
d
.
T
h
e
fea
t
u
re
e
x
t
rac
t
i
o
n
me
t
h
o
d
s
u
s
ed
i
n
t
h
i
s
p
ap
er
i
s
G
eo
me
t
ri
c
mo
men
t
s
an
d
Co
l
o
r
feat
u
re
ex
t
rac
t
i
o
n
.
T
h
e
reco
g
n
i
t
i
o
n
rat
e
o
f
9
5
.
5
5
6
%
h
a
s
b
een
ac
h
i
e
v
ed
.
T
h
e
ex
p
eri
me
n
t
a
l
res
u
l
t
i
l
l
u
s
t
ra
t
i
o
n
s
t
h
at
t
h
e
as
s
o
c
i
at
i
o
n
o
f
t
w
o
t
ech
n
i
q
u
e
s
t
h
a
t
p
r
o
v
i
d
e
b
e
t
t
er
acc
u
racy
.
K
e
y
w
o
r
d
s
:
F
a
c
e
r
e
c
ognit
ion
F
e
a
tur
e
ba
s
e
d
F
uz
z
y
Ne
ur
a
l
n
e
twor
k
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
t
h
or
:
Ha
ms
a
A.
Abdullah,
C
oll
e
ge
of
I
n
f
or
mation
E
nginee
r
ing
,
Al
-
Na
hr
a
in
Unive
r
s
it
y,
I
r
a
q
.
E
mail:
ha
ms
a
.
a
bdulkar
e
e
m@c
oie
-
na
hr
a
in.
e
du.
iq
1.
I
NT
RODU
C
T
I
ON
R
e
c
e
ntl
y,
human
f
a
c
e
r
e
c
ogna
ti
on
is
a
n
im
por
tant
r
e
s
e
a
r
c
h
topi
c
in
the
f
ields
of
a
r
ti
f
icia
l
int
e
ll
igenc
e
a
nd
pa
tt
e
r
n
identif
ica
ti
on
.
HFR
ha
s
s
e
ve
r
a
l
is
s
ue
s
uc
h
a
s
:
ha
i
r
a
nd
e
xp
r
e
s
s
ions
c
a
n
c
ha
nge
the
f
a
c
e
;
s
im
il
a
r
it
y
be
twe
e
n
dif
f
e
r
e
nt
f
a
c
e
s
;
a
nd
a
ls
o
ther
e
a
r
e
dif
f
e
r
e
nt
a
ngles
the
f
a
c
e
c
a
n
be
view
e
d.
A
good
HFR
s
ys
tem
mus
t
be
r
obus
t
to
ove
r
c
ome
thes
e
is
s
ue
s
[
1]
.
HFR
s
ys
tem
divi
de
d
int
o
thr
e
e
s
tage
s
:
d
e
tec
ti
on;
f
e
a
tur
e
e
xtr
a
c
ti
on;
a
nd
r
e
c
ognit
ion
[
2,
3]
.
A
r
ti
f
icia
l
ne
ur
a
l
ne
twor
ks
(
AN
N)
we
r
e
us
e
d
wid
e
ly
f
or
c
ons
tr
uc
ti
ng
int
e
ll
igent
c
omput
e
r
s
y
s
tems
ba
s
e
d
on
im
a
ge
pr
oc
e
s
s
ing
a
nd
pa
tt
e
r
n
r
e
c
ognit
i
on
[
4
]
.
T
he
ba
c
kpr
opa
ga
ti
on
ne
ur
a
l
ne
twor
k
(
B
P
NN
)
is
t
he
mos
t
c
omm
on
AN
N
model
that
c
a
n
be
tr
a
ined
us
ing
B
P
a
lgor
it
hm
[
5]
.
A
lot
of
s
tudi
e
s
a
bou
t
HF
R
s
ys
tem
,
e
a
c
h
one
of
them
de
pe
nds
on
dif
f
e
r
e
nt
methods
s
uc
h
a
s
:
e
igen
va
lues
of
f
a
c
e
,
f
e
a
tur
e
s
,
gr
a
ph
m
a
tching
,
m
a
tching
of
t
e
mpl
a
te
,
a
nd
AN
N
methods
[
6]
.
I
n
[
7]
,
Ne
ur
o
-
f
uz
z
y
(
NF)
f
us
ion
in
a
mul
ti
moda
l
f
a
c
e
r
e
c
ognit
ion
us
ing
P
C
A,
I
C
A
a
nd
S
I
F
T
is
int
r
oduc
e
d.
I
n
thi
s
wor
k
,
mul
t
im
oda
l
f
a
c
e
r
e
c
ognit
ion
is
dis
c
us
s
e
d
a
nd
the
im
pleme
nted
of
with
NF
c
ombi
na
ti
on.
T
he
pr
incipa
l
c
omponent
a
na
lys
is
(
P
C
A)
a
nd
indepe
nde
nt
c
omponent
a
na
lys
is
(
I
C
A)
a
s
we
ll
a
s
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
.
1
,
F
e
br
ua
r
y
2020
:
427
-
435
428
fe
a
tur
e
e
xtr
a
c
ti
on
ba
s
e
d
on
S
I
F
T
a
r
e
us
e
d.
T
he
r
e
c
ognit
ion
I
D
de
ter
mi
ne
ba
s
e
d
on
NF
inf
e
r
e
nc
e
s
ys
tem.
I
n
[
8]
,
f
a
c
e
r
e
c
ognit
ion
us
ing
ne
ur
o
-
f
uz
z
y
a
nd
e
igenf
a
c
e
i
s
int
r
oduc
e
d.
I
n
thi
s
wor
k
,
a
human
pr
e
s
e
nc
e
is
de
t
e
c
ted
by
e
xtr
a
c
ti
ng
the
s
kin
a
r
e
a
by
us
ing
the
E
igen
va
lu
e
of
f
a
c
e
method
.
T
he
n
bu
us
ing
a
ne
ur
o
-
f
uz
z
y
method,
the
f
a
c
e
is
r
e
c
ognize
d.
I
n
[
9]
,
f
a
c
e
r
e
c
ognit
ion
s
ys
tem
us
ing
a
da
pti
ve
ne
ur
o
f
uz
z
y
inf
e
r
e
nc
e
s
ys
tem
(
AN
F
I
S
)
is
int
r
oduc
e
d.
I
n
thi
s
wor
k
,
AN
F
I
S
wi
th
P
C
A
a
lgor
it
h
m
ha
s
be
e
n
pr
opos
e
d
by
c
ons
id
e
r
ing
dif
f
e
r
e
nt
c
ontr
ibut
ions
of
the
t
r
a
ini
ng
s
a
mpl
e
s
.
I
n
[
10]
f
a
c
e
r
e
c
ognit
ion
(
F
R
)
ba
s
e
d
on
de
s
c
ion
leve
l
f
us
ion
is
int
r
oduc
e
d.
I
n
t
his
pa
pe
r
,
a
ne
w
method
na
med
C
2D
C
NN
i
s
pr
opos
e
d.
I
n
[
1
1]
f
a
c
ial
r
e
c
ognit
ion
ba
s
e
d
o
n
a
da
pti
ve
ne
u
r
o
f
uz
z
y
(
AN
F
)
inf
e
r
e
nc
e
s
ys
tem
is
int
r
oduc
e
d
.
I
n
thi
s
wor
k,
th
e
manin
c
ontr
ibut
ion
is
ba
s
e
d
on
f
e
a
tur
e
e
xtr
a
c
ti
on
a
nd
c
las
s
if
ica
ti
on.
T
he
a
im
of
thi
s
pa
pe
r
is
to
de
ve
lop
a
HFR
sy
s
tem
ba
s
e
d
on
f
e
tur
e
e
xt
r
a
c
ti
on
b
y
us
ing
Ne
ur
o
-
F
uz
z
y
I
nter
f
e
r
a
nc
e
s
ys
tem
.
T
he
p
r
opos
e
d
s
ys
tem
c
ons
is
t
of
two
s
tage
s
:
f
ir
s
t
s
tage
f
a
c
e
r
e
c
ognit
ion
by
us
ing
NN
a
nd
s
e
c
ond
s
tage
is
to
e
va
luate
the
pe
r
f
o
r
manc
e
of
the
pr
opos
e
d
a
lgor
it
hm
with
f
uz
z
y
s
ys
tem.
2.
F
AC
E
RE
COGNI
T
I
ON
T
E
CHNI
QUE
S
T
he
main
s
teps
to
f
a
c
e
r
e
c
ognit
ion
a
r
e
;
e
xtr
a
c
ti
ng
t
he
f
e
a
tur
e
s
f
r
om
the
im
a
ge
s
,
s
tor
e
f
e
a
tur
e
s
in
da
ta
ba
s
e
,
de
s
ign
NN,
tr
a
in
f
e
a
tur
e
on
ne
twor
k
,
a
nd
tes
t
the
old
a
nd
ne
w
da
ta
NN
.
2
.
1
.
F
ac
e
f
e
at
u
r
e
e
x
t
r
ac
t
ion
wi
t
h
m
om
e
n
t
s
F
e
a
tur
e
e
xtr
a
c
ti
on
is
a
s
e
c
ti
on
o
f
pa
tt
e
r
n
r
e
c
ognit
ion
tec
hniques
whic
h
int
e
nt
to
e
xtr
a
c
t
o
r
r
e
tr
ieve
the
indi
vidual
va
lues
f
r
om
a
n
objec
t
that
dif
f
e
r
e
n
ti
a
tes
it
f
r
o
m
other
objec
ts
[
12]
.
F
e
a
tur
e
e
xtr
a
c
ti
o
n
f
or
a
n
im
a
ge
c
a
n
be
done
by
us
ing
s
e
ve
r
a
l
methods
s
uc
h
a
s
invar
iant
mom
e
nts
a
nd
c
olo
r
f
e
a
tur
e
e
xt
r
a
c
ti
on.
I
n
im
a
ge
a
na
lys
is
a
ppli
c
a
ti
ons
,
the
I
mage
int
r
oduc
e
e
f
f
e
c
t
ive
de
s
c
r
ip
ti
on
.
T
he
main
a
dva
ntage
of
us
ing
i
mage
f
or
a
na
lys
is
a
ppli
c
a
ti
on
is
their
c
a
pa
bil
it
y
to
int
r
oduc
e
in
va
r
iant
mea
s
ur
e
s
of
s
ha
pe
[
13]
.
M
oment
ba
s
e
d
f
e
a
tur
e
de
s
c
r
ipt
ion
ha
ve
de
ve
loped
int
o
a
e
xtr
ode
na
r
y
tool
f
or
im
a
ge
a
na
lys
is
a
ppli
c
a
ti
ons
[
14]
.
2
.
1
.
1
.
Geom
e
t
r
ic
m
om
e
n
t
s
(
GM)
GM
pr
ov
ed
to
be
a
n
e
f
f
e
c
ti
ve
a
na
l
ys
is
method
f
or
im
a
ge
a
ppli
c
a
ti
on
.
GM
c
a
n
be
us
e
d
f
or
di
f
f
e
r
e
nt
a
ppli
c
a
ti
on
s
uc
h
a
s
:
a
ir
c
r
a
f
t
identif
ica
ti
on,
c
ha
r
a
c
ter
r
e
c
ognit
ion
,
s
ha
pe
a
nd
im
a
ge
a
na
lys
is
,
nor
maliza
ti
on
of
im
a
ge
,
c
olor
textur
e
r
e
c
ognit
ion
,
de
tec
ti
on
of
a
c
c
ur
a
te
pos
it
ion,
r
e
tr
ieva
l
o
f
i
mage
a
nd
many
types
of
im
a
ge
pr
oc
e
s
s
ing
a
ppli
c
a
ti
ons
.
F
or
a
2D
de
ns
it
y
f
unc
ti
on
p(
x,
y)
,
the
(
p+
q)
th
or
de
r
GM
m
pq
a
r
e
de
f
ined
by
[
1
5,
16]
:
=
∑
∑
=
1
=
1
(
,
)
(
1
)
2
.
1
.
2
.
Color
m
om
e
n
t
s
(
CM
)
CM
a
r
e
mea
s
ur
e
ments
whic
h
can
be
us
e
d
f
or
c
ontr
a
s
ti
ng
im
a
ge
s
a
c
c
or
ding
to
c
olo
r
f
e
a
tur
e
s
of
the
im
a
ge
s
.
T
he
ba
s
is
o
f
CM
is
ba
s
e
d
on
the
hypoth
e
s
is
that
the
c
olo
r
d
is
tr
ibu
ti
on
in
a
n
im
a
ge
c
a
n
be
e
xplaine
d
as
the
dis
tr
ibut
ion
of
pr
oba
bil
it
ies
(
P
D)
.
P
D
a
r
e
notable
by
a
number
of
unique
mom
e
nts
[
17]
.
T
he
f
ir
s
t
mom
e
nt
is
mea
n,
the
s
e
c
ond
is
s
tanda
r
d
de
viation,
a
nd
the
las
t
one
is
s
ke
wn
e
s
s
.
T
he
CM
ha
ve
be
e
n
de
mons
tr
a
ted
to
be
e
f
f
icie
nt
in
r
e
pr
e
s
e
nti
ng
im
a
ge
s
c
olor
dis
tr
ibut
ions
[
18]
.
a.
M
oment
1
-
mea
n
=
1
∑
∑
(
,
)
=
1
=
1
(
2
)
b.
M
oment
2
-
S
tanda
r
d
De
viation;
=
(
1
∑
∑
(
(
,
=
1
=
1
)
-
)
2
)
(
3
)
c.
M
oment
3
-
S
ke
wne
s
s
=
(
1
∑
∑
(
(
,
=
1
=
1
)
-
)
3
)
1
3
⁄
(
4
)
w
he
r
e
(
,
)
is
the
im
a
ge
pixel
a
nd
M
,
N
r
e
pr
e
s
e
nt
the
h
ight
a
nd
wie
dit
h
of
the
im
a
ge
r
e
s
pe
c
ti
ve
ly
[
19
]
.
2
.
2
.
T
h
e
NN
t
e
c
h
n
iq
u
e
s
T
he
goa
l
of
the
NN
tec
hnique
is
dis
ti
nguis
h
a
hu
man
f
a
c
e
(
HF)
by
tr
a
in
ing
the
ne
two
r
k.
I
n
the
NN
tec
hnique,
ther
e
a
r
e
two
pha
s
e
s
to
r
e
c
ognize
a
HF
.
T
he
f
ir
s
t
pha
s
e
is
the
t
r
a
ini
ng
a
nd
the
s
e
c
ond
pha
s
e
is
the
tes
ti
ng.
I
n
the
t
r
a
ini
ng
pha
s
e
,
the
s
e
t
of
tr
a
ini
ng
da
ta
c
ontain
input
s
e
t
up
with
it
’
s
out
put
a
s
a
c
ons
e
que
nc
e
.
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
N
e
ur
o
-
fuz
z
y
inf
e
r
e
nc
e
s
y
s
tem
bas
e
d
face
r
e
c
ognit
i
on
us
ing
featur
e
e
x
tr
ac
ti
on
(
Ham
s
a
A
.
A
bdu
ll
ah)
429
T
he
n
the
NN
is
t
r
a
in
ing
ba
s
e
d
on
the
da
ta
to
r
e
gulate
thr
e
s
holds
a
nd
we
ight
s
of
the
ne
twor
k's
to
r
e
duc
e
pr
e
dictions
e
r
r
or
[
20]
.
2
.
2
.
1
.
B
ac
k
-
p
r
op
agat
ion
n
e
u
r
al
n
e
t
wor
k
(
B
P
NN
)
B
P
NN
ha
s
thr
e
e
main
laye
r
e
d
s
tr
uc
tur
e
whic
h
a
r
e
:
t
he
input
laye
r
s
,
hidden
a
nd
output
laye
r
s
.
I
n
B
P
NN
the
we
ight
s
of
the
ne
twor
k's
a
r
e
ob
taine
d
thr
ough
lea
r
n
ing.
T
he
c
ompl
e
xit
y
of
t
r
a
ini
ng
de
p
e
nds
on
the
number
of
the
hidden
laye
r
s
,
whe
r
e
the
c
ompl
e
xit
y
of
tr
a
ini
ng
incr
e
a
s
e
with
incr
e
a
s
ing
of
hidd
e
n
laye
r
number
.
T
he
B
P
NN
tr
a
ini
ng
is
a
c
hieve
d
in
thr
e
e
s
teps
[
21,
22]
whic
h
a
r
e
input
f
e
e
d
-
f
or
wa
r
d
(
F
F
)
,
e
r
r
or
a
nd
we
ight
s
c
omput
a
ti
on
a
nd
e
r
r
o
r
ba
c
k
-
pr
opa
ga
ti
on
(
E
B
P
)
.
T
he
e
xter
na
l
inpu
ts
f
e
ds
the
unit
s
of
t
he
input
lay
e
r
wi
thout
c
onne
c
ti
on
int
o
a
laye
r
.
T
he
n
the
f
i
r
s
t
hidden
laye
r
f
e
ds
by
the
input
laye
r
.
T
he
hi
dde
n
laye
r
r
e
c
e
ives
a
we
ight
e
d
bias
a
f
ter
im
pleme
nts
the
a
c
ti
va
ti
on
f
unc
ti
on.
T
he
ne
xt
hidden
laye
r
f
e
ds
by
the
output
of
the
pr
e
vious
ly
hidden
laye
r
.
T
h
is
pr
oc
e
dur
e
c
a
r
r
ies
on
ti
ll
the
las
t
hidden
laye
r
.
T
he
output
la
ye
r
f
e
ds
by
the
output
s
of
the
las
t
hidden
laye
r
.
Although
the
tr
a
ini
ng
o
f
B
P
NN
is
s
o
s
low,
the
mom
e
nt
that
t
he
NN
is
tr
a
ined,
it
f
ulf
il
ls
the
r
e
s
ult
s
quickly
[
15]
.
2
.
3
.
F
u
z
z
y
logi
c
(
F
L
)
FL
is
a
method
o
f
log
ic
a
l
va
lue
.
I
t
c
onc
e
r
ning
of
r
e
a
s
oning
that
is
c
onve
r
ge
nt
ins
ted
of
e
xa
c
t
a
nd
f
ixed.
T
r
a
dit
ional
ly,
the
binar
y
ga
ther
s
t
r
ue
o
r
f
a
ls
e
va
l
ue
s
.
T
he
r
a
nge
s
of
tr
u
e
va
lue
is
be
twe
e
n
0
a
nd
1.
I
n
F
L
the
idea
of
f
r
a
c
ti
ona
l
t
r
uth
ha
s
be
e
n
us
e
d
,
whe
r
e
t
he
r
a
nge
c
ould
be
f
u
ll
y
f
a
ls
e
a
nd
f
ull
y
t
r
ue
.
M
or
e
ove
r
,
a
s
pe
c
if
ic
f
unc
ti
ons
may
be
us
e
d
to
mana
ge
li
nguis
ti
c
va
r
iable
.
I
r
r
a
ti
ona
li
ty
c
a
n
be
de
s
c
r
ibed
in
ter
ms
o
f
wha
t
is
known
a
s
the
f
uz
z
jec
ti
ve
[
23
-
25]
3.
T
HE
P
ROP
OS
E
D
AL
GO
RI
T
HM
R
e
c
e
ntaly,
many
a
lgo
r
it
hms
f
or
HFR
a
r
e
in
tr
oduc
e
d.
I
n
thes
e
a
lgor
it
hms
,
the
p
r
oc
e
s
s
ing
r
e
late
d
to
the
type
of
im
a
ge
or
f
e
a
tur
e
e
xtr
a
c
ti
on
is
us
e
d
mos
tl
y.
I
n
thi
s
pa
pe
r
,
a
ll
type
s
of
im
a
ge
s
c
a
n
be
us
e
d
a
s
input
s
to
the
pr
opos
e
d
a
lgor
it
hm
.
T
he
p
r
opos
e
d
a
lgor
it
hm
is
c
ons
is
t
of
two
s
tage
s
:
the
tr
a
ini
ng
a
nd
r
e
c
ognit
ion
s
tage
a
nd
the
FL
s
tage
.
3
.
1
.
T
r
ain
in
g
an
d
r
e
c
ogn
at
ion
o
f
t
h
e
NN
F
igur
e
1
s
hows
that
the
NN
t
r
a
ini
ng
c
ons
is
ts
of
th
r
e
e
s
tage
s
.
I
n
the
f
i
r
s
t
s
tage
,
the
s
e
t
of
im
a
ge
s
a
r
e
tr
a
ini
ng
to
s
upply
the
da
ta
to
ne
twor
k
.
T
he
r
e
f
or
e
,
the
de
s
igni
ng
s
tr
uc
tur
e
of
input
r
e
quir
e
d
the
ident
ica
l
r
ow
f
r
om
the
im
a
ge
matr
ix
a
s
s
hown
in
F
ig
ur
e
2.
I
n
th
e
pr
opos
e
d
a
lgor
i
thm
,
B
P
NN
ha
s
a
laye
r
e
d
s
tr
uc
tur
e
a
s
:
18
,
37,
5
.
T
he
s
e
laye
r
s
a
r
e
input
,
hidden
a
nd
outpu
t
la
ye
r
.
T
he
hidden
l
a
ye
r
in
the
ne
twor
k
c
a
n
be
mor
e
t
ha
n
one
,
but
one
laye
r
is
a
ppr
opr
iate
to
r
e
a
c
h
ou
r
goa
l
.
F
igu
r
e
2
s
hows
the
de
s
igned
NN
a
r
c
hit
e
c
tur
e
.
T
he
de
s
i
gne
d
NN
ne
twor
k
is
tr
a
ined
unti
l
the
output
s
of
NN
a
r
e
e
qua
l
to
de
s
ir
e
d
output
s
.
T
he
output
r
e
s
ult
s
of
the
p
r
op
os
e
d
NN
be
c
ome
mor
e
a
c
c
ur
a
te
whe
n
the
tr
a
ined
va
lue
be
c
ome
matc
hing
to
the
de
s
ir
e
d
output
.
S
o,
the
pr
op
os
e
d
NN
c
a
n
be
tr
a
ined
up
to
4
to
5
ti
mes
to
ge
t
the
de
s
ir
e
d
output
s
.
F
igur
e
1
.
T
r
a
ini
ng
o
f
n
e
ur
a
l
n
e
twor
k
F
igur
e
2.
De
s
ign
the
ne
ur
a
l
ne
twor
k
a
r
c
hit
e
c
tur
e
F
igur
e
3
s
hows
the
block
diagr
a
m
t
r
a
ini
ng
a
nd
tes
ti
ng
o
f
NN
.
I
n
thi
s
f
igu
r
e
,
ther
e
a
r
e
two
s
tage
s
in
t
he
s
ys
tem:
tr
a
ini
ng
a
nd
tes
ti
ng
s
tage
.
I
n
the
tr
a
in
ing
s
tage
,
the
s
e
ts
of
tr
a
ini
ng
im
a
ge
us
e
d
to
t
r
a
in
ing
NN
.
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
.
1
,
F
e
br
ua
r
y
2020
:
427
-
435
430
T
he
f
ir
s
t
s
tep
in
the
t
r
a
ini
ng
pha
s
e
is
f
e
a
tur
e
e
xtr
a
c
ti
on.
W
he
r
e
the
GM
a
nd
C
M
a
r
e
s
ue
d
to
e
xtr
a
c
t
18
f
e
a
tur
e
s
f
or
m
e
a
c
h
s
ingl
e
t
r
a
in
im
a
ge
.
T
he
n
thes
e
f
e
a
tur
e
s
s
t
or
e
d
in
the
da
taba
s
e
of
the
s
ys
tem
to
be
us
e
d
in
the
tr
a
ini
ng
s
tage
.
T
he
NN
de
s
igned
a
c
c
or
ding
to
the
s
e
t
of
i
nput
da
ta
a
nd
the
s
ize
o
f
the
de
s
ir
e
d
output
s
.
Af
ter
that
,
the
de
s
igned
NN
will
be
tr
a
ini
ng
to
r
e
c
ognize
the
im
a
ge
s
tor
e
d
in
th
e
da
taba
s
e
of
the
s
ys
tem.
I
n
th
e
tes
ti
ng
s
tage
,
the
s
e
ts
of
tes
t
i
mage
that
us
e
d
to
tes
t
the
a
c
c
ur
a
c
y
of
the
de
s
igned
NN
.
T
he
f
ir
s
t
s
tep
in
the
tes
ti
ng
pha
s
e
is
the
f
e
utur
e
e
xt
r
a
c
ti
on
by
us
ing
GM
a
nd
C
M
.
T
h
e
n
the
e
xt
r
a
c
ted
f
e
a
tur
e
s
us
e
d
i
n
tes
ti
ng
s
tage
to
f
in
d
in
they
matc
h
with
f
e
a
tur
e
that
s
tor
e
d
in
da
taba
s
e
of
the
s
ys
tem.
F
igur
e
3.
T
r
a
ini
ng
a
nd
tes
ti
ng
of
the
ne
ur
a
l
ne
twor
k
3
.
2
.
F
u
z
z
y
logi
c
T
wo
pa
r
a
mete
r
s
a
r
e
us
e
d
to
de
f
ine
the
a
c
c
ur
a
c
y
o
f
the
de
s
igned
NN
whic
h
a
r
e
Gr
a
dient
a
nd
E
poc
hs
.
T
he
gr
a
dient
is
the
opt
im
iza
ti
on
method
us
e
d
f
o
r
the
lea
r
ini
ng
s
ys
tem.
Gr
a
dient
pa
r
a
mete
r
s
r
e
f
e
r
to
f
ind
the
s
lope
of
e
r
r
or
a
nd
de
c
r
e
a
s
ing
the
e
r
r
or
s
lop
by
modi
f
ying
the
we
i
ghts
a
nd
bias
unti
l
mi
ni
mi
z
ing
the
leve
l
of
e
r
r
o
r
.
E
poc
hs
pa
r
a
mete
r
r
e
f
e
r
s
to
the
numbe
r
o
f
ti
me
that
the
a
lgor
it
h
m
p
r
oc
e
s
s
e
s
the
da
tas
e
t.
I
n
E
poc
hs
pa
r
a
mete
r
thr
e
e
ter
ms
a
r
e
ve
r
y
s
igni
f
ica
nt
whic
h
a
r
e
:
the
number
of
e
poc
hs
,
ti
me
,
a
nd
va
lue.
T
he
ne
twor
k
le
a
r
ns
in
s
li
ght
r
e
it
e
r
a
ti
ons
whe
n
the
number
of
e
p
oc
hs
is
f
e
we
r
.
T
he
ti
me
in
e
poc
hs
r
e
f
e
r
s
to
the
ne
t
wor
k
to
r
e
a
c
h
it
s
objec
ti
ve
s
hor
tl
y
a
nd
e
a
s
il
y.
And
whe
n
t
he
va
lue
of
e
poc
hs
is
low
that
mea
ns
the
ne
twor
k
is
higher
a
c
c
ur
a
c
y.
T
he
ne
ur
o
f
uz
z
y
b
lock
diagr
a
m
is
s
hown
in
F
igur
e
4.
F
igur
e
4
s
hows
that
the
two
outpu
t
pa
r
a
mete
r
s
(
E
poc
hs
a
nd
Gr
a
dient)
of
NN
is
f
e
e
d
F
I
S
s
uc
h
a
s
i
nputs
that
a
c
c
or
dingl
y
o
f
whic
h
a
c
c
ur
a
c
y
is
c
ons
id
e
r
e
d.
F
igur
e
4.
Ne
ur
o
f
uz
z
y
b
lock
d
iag
r
a
m
T
he
F
I
S
input
s
is
int
r
oduc
e
d
thr
ough
th
e
membe
r
s
hip
f
unc
ti
on
(
M
F
)
that
is
a
ll
oc
a
te
the
pos
it
ion
of
two
input
va
lues
li
e
in
r
a
nge
of
va
lues
.
T
he
r
a
nge
o
f
e
poc
hs
that
us
e
d
in
thi
s
pa
pe
r
is
s
e
lec
ted
a
s
0
to
7
0.
W
hil
e
the
r
a
nge
of
gr
a
dient
that
us
e
d
in
thi
s
pa
pe
r
is
s
e
le
c
ted
a
s
0
to
0.
8.
A
ny
membe
r
s
hip
f
unc
ti
on
c
a
n
be
s
e
lec
t
ed
f
r
om
whic
h
a
r
e
a
l
r
e
a
dy
c
us
tom
ize
d
or
de
f
ined
me
mber
s
hip.
F
igur
e
5
the
de
s
igned
M
F
of
Gr
a
dient
a
n
d
E
poc
hs
input
s
by
us
ing
M
a
tl
a
b
.
W
hil
e
F
ig
ur
e
6
s
hows
the
de
s
igned
M
F
of
Gr
a
dient
a
nd
E
poc
hs
input
s
by
us
ing
F
I
S
.
F
igur
e
7
s
hows
t
he
M
F
of
the
output
va
r
iable
(
a
c
c
ur
a
c
y)
.
T
he
r
a
nge
of
of
a
c
c
ur
a
c
y
us
e
d
in
thi
s
pa
pe
r
is
s
e
lec
ted
as
0
to
100.
W
he
n
the
Gr
a
dient
a
nd
E
poc
hs
inp
uts
a
r
e
low
in
r
a
nge
,
the
a
c
c
ur
a
c
y
is
s
e
t
up
a
c
c
or
ding
to
the
r
ules
.
T
h
e
s
e
r
ules
a
r
e
ba
s
e
d
on
the
a
malga
m
a
ti
on
of
Gr
a
dient
a
nd
E
poc
hs
input
pa
r
a
mete
r
s
t
o
f
ulf
il
l
the
de
s
ir
e
d
output
a
s
s
hown
in
F
igur
e
8.
T
he
f
uz
z
y
r
ules
a
r
e
ne
c
e
s
s
a
r
y
f
or
e
a
c
h
input
,
the
s
e
t
o
f
r
ules
mus
t
be
r
e
pe
a
ted
f
or
the
input
pa
r
a
mete
r
.
T
he
r
ule
in
F
igu
r
e
8
(
a
)
is
im
pleme
nted
by
us
ing
M
a
tl
a
b
c
od
e
while
the
r
ules
in
(
b)
a
r
e
im
pleme
nted
by
us
ing
F
I
S
.
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
N
e
ur
o
-
fuz
z
y
inf
e
r
e
nc
e
s
y
s
tem
bas
e
d
face
r
e
c
ognit
i
on
us
ing
featur
e
e
x
tr
ac
ti
on
(
Ham
s
a
A
.
A
bdu
ll
ah)
431
F
igur
e
5
.
M
e
mber
s
hip
f
unc
ti
on
f
or
input
gr
a
dient
a
nd
e
poc
h
by
us
ing
m
t
lab
c
ode
(
a
)
(
b)
F
igur
e
6
.
M
e
mber
s
hip
f
unc
ti
on
by
us
ing
F
I
S
f
unc
ti
on
:
(
a
)
f
or
i
nput
g
r
a
dient
,
(
b)
i
nput
e
poc
hs
(
a
)
(
b)
F
igur
e
7
.
Output
m
e
mber
s
hip
f
unc
ti
on
:
(
a
)
by
us
in
g
F
I
S
f
unc
ti
on
,
(
b
)
by
us
ing
m
tl
a
b
c
ode
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
.
1
,
F
e
br
ua
r
y
2020
:
427
-
435
432
(
a
)
(
b)
F
igur
e
8
.
T
he
r
ules
f
o
r
the
ne
twor
k
,
(
a
)
is
im
pleme
nted
by
us
ing
matlab
c
ode
while
the
r
ules
in
(
b)
a
r
e
im
pleme
nted
by
us
ing
F
I
S
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
N
e
ur
o
-
fuz
z
y
inf
e
r
e
nc
e
s
y
s
tem
bas
e
d
face
r
e
c
ognit
i
on
us
ing
featur
e
e
x
tr
ac
ti
on
(
Ham
s
a
A
.
A
bdu
ll
ah)
433
4.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
he
r
e
a
r
e
two
s
tage
s
of
the
r
e
s
ult
s
of
the
pr
opos
e
d
a
lgor
it
hm
:
ta
r
ning
a
nd
tes
ti
ng
s
tage
.
I
n
the
tr
a
in
ing
s
tage
,
human
f
a
c
e
will
be
r
e
c
ognize
d
by
the
pr
opos
e
d
s
ys
tem
while
the
te
s
t
s
e
t
of
im
a
ge
c
a
nnot
be
r
e
c
ognize
d
by
the
pr
opos
e
d
s
ys
tem
be
c
a
us
e
they
a
r
e
not
in
the
da
taba
s
e
of
the
s
ys
tem
a
s
s
hown
in
F
igur
e
9
.
T
he
r
e
s
ult
s
of
th
is
s
tage
is
the
output
of
ne
ur
a
l
ne
twor
k
whic
h
is
e
poc
h
a
nd
gr
a
dient
.
W
he
n
the
number
of
the
e
poc
h
is
mi
nim
um
,
that
mea
ns
the
s
ys
tem
ge
ts
the
tar
ge
t
output
with
mi
ni
mum
it
e
r
a
ti
on
that
lea
d
to
de
c
r
e
a
s
e
the
ti
me
o
f
tr
a
ini
ng.
Als
o,
whe
n
the
va
lue
of
gr
a
dient
is
m
ini
mum
,
thi
s
mea
ns
the
s
ys
tem
is
lea
r
ning
with
de
c
r
e
s
ing
the
e
r
r
or
s
lop
by
modi
f
y
the
we
igh
ts
a
nd
bias
unti
l
mi
nim
izing
the
leve
l
of
e
r
r
or
.
T
he
number
of
f
e
a
tur
e
that
e
x
tr
a
c
t
f
r
om
f
a
c
e
im
a
ge
a
nd
us
e
d
a
s
a
da
taba
s
e
c
a
n
be
incr
e
a
s
e
d
to
e
nha
nc
e
the
a
c
c
ur
a
c
y
of
the
pr
opos
e
d
s
ys
tem.
F
igur
e
10
s
hows
the
tr
a
ini
ng
pe
r
f
or
manc
e
of
the
pr
opos
e
d
s
ys
tem.
T
he
f
igur
e
s
hows
that
the
tr
a
ini
ng
c
ur
ve
is
r
e
a
c
he
d
to
it
s
tar
ge
ts
th
r
ough
modi
f
ica
ti
on
of
bias
e
s
a
nd
we
ight
s
.
I
n
thi
s
pa
pe
r
,
two
s
e
ts
of
the
f
a
c
e
im
a
ge
a
r
e
us
e
d
I
n
or
de
r
to
e
va
luate
the
pe
r
f
or
manc
e
of
the
p
r
opos
e
d
s
ys
tem.
T
he
f
ir
s
t
s
e
t
is
the
tr
a
ini
ng
im
a
ge
a
nd
the
s
e
c
ond
s
tep
is
the
tes
ti
ng
im
a
ge
.
T
he
pe
r
f
o
r
manc
e
e
va
luation
of
the
pr
opos
e
d
s
ys
tem
is
done
by
us
ing
thes
e
tow
type
of
s
e
ts
f
a
c
e
im
a
g
e
.
T
he
tes
t
im
a
ge
s
a
r
e
us
e
d
to
the
tr
a
ined
NN
to
d
e
ter
mi
ne
the
pe
r
c
e
ntage
of
e
r
r
o
r
a
nd
a
c
c
ur
a
c
y
of
the
pr
op
os
e
d
s
ys
tem.
A
30
s
e
t
of
f
a
c
e
im
a
ge
s
a
r
e
us
e
d
a
s
tr
a
ini
ng
im
a
ge
s
a
nd
a
s
e
t
of
15
f
a
c
e
im
a
ge
s
a
r
e
us
e
d
a
s
a
t
e
s
ti
ng
im
a
ge
.
T
a
ble
1
s
hows
that
the
r
e
c
ognit
ion
r
e
s
ult
s
of
the
4
5
f
a
c
e
im
a
ge
s
.
T
he
s
e
r
e
s
ult
s
a
r
e
a
na
lyze
d
to
f
ind
the
r
e
c
ognit
ion
r
a
te
(
R
T
)
o
f
the
pr
opos
e
d
s
ys
tem
a
s
s
hown
in
T
a
ble
2
.
A
r
e
c
ognit
ion
r
a
te
of
95
.
556%
i
s
c
a
lcula
ted
f
o
r
the
pr
opos
e
d
s
ys
tem.
T
his
R
T
va
lu
e
is
quit
e
a
ppr
opr
iate
f
or
f
a
c
e
r
e
c
ognit
ion
s
ys
tems
.
Th
e
s
e
c
ond
s
tage
of
the
r
e
s
ult
s
,
is
to
c
a
lcula
te
the
pe
r
f
o
r
manc
e
of
the
pr
opos
e
d
a
lgor
it
hm
wi
th
F
uz
z
y
s
ys
tem.
T
he
o
utput
of
the
Ne
ur
a
l
Ne
twor
k
s
ys
tem
(
e
poc
h
a
nd
g
r
a
dien
t)
us
e
d
a
s
input
to
F
uz
z
y
S
ys
tem
to
identi
f
y
the
a
c
c
ur
a
c
y
of
the
s
ys
tem
a
s
s
hown
in
F
igur
e
11
.
(
a
)
(
b)
F
igur
e
9
.
R
e
c
ogniza
ti
on
of
the
tes
t
a
nd
t
r
a
ined
pe
r
s
on
:
(
a
)
tr
a
ined
i
mage
,
(
b
)
t
e
s
ted
i
mage
F
igur
e
10
.
T
r
a
ini
ng
p
e
r
f
or
manc
e
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
.
1
,
F
e
br
ua
r
y
2020
:
427
-
435
434
T
a
ble
1
.
R
e
c
ogn
a
ti
on
r
e
s
ult
s
I
ma
ge
N
o.
T
ype
R
e
s
ul
ts
I
ma
ge
N
o.
T
ype
R
e
s
ul
ts
1
T
r
a
in
e
R
e
c
ogni
s
e
d
24
T
r
a
in
e
R
e
c
ogni
s
e
d
2
T
r
a
in
e
R
e
c
ogni
s
e
d
25
T
r
a
in
e
R
e
c
ogni
s
e
d
3
T
r
a
in
e
R
e
c
ogni
s
e
d
26
T
r
a
in
e
R
e
c
ogni
s
e
d
4
T
r
a
in
e
R
e
c
ogni
s
e
d
27
T
r
a
in
e
R
e
c
ogni
s
e
d
5
T
r
a
in
e
R
e
c
ogni
s
e
d
28
T
r
a
i
ne
R
e
c
ogni
s
e
d
6
T
r
a
in
e
R
e
c
ogni
s
e
d
29
T
r
a
in
e
R
e
c
ogni
s
e
d
7
T
r
a
in
e
R
e
c
ogni
s
e
d
30
T
r
a
in
e
R
e
c
ogni
s
e
d
8
T
r
a
in
e
R
e
c
ogni
s
e
d
31
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
9
T
r
a
in
e
R
e
c
ogni
s
e
d
32
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
10
T
r
a
in
e
R
e
c
ogni
s
e
d
33
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
11
T
r
a
in
e
R
e
c
ogni
s
e
d
34
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
12
T
r
a
in
e
R
e
c
ogni
s
e
d
35
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
13
T
r
a
in
e
R
e
c
ogni
s
e
d
36
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
14
T
r
a
in
e
R
e
c
ogni
s
e
d
37
T
e
s
t
R
e
c
ogni
s
e
d
15
T
r
a
in
e
R
e
c
ogni
s
e
d
38
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
16
T
r
a
in
e
R
e
c
ogni
s
e
d
39
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
17
T
r
a
in
e
R
e
c
o
gni
s
e
d
40
T
e
s
t
R
e
c
ogni
s
e
d
18
T
r
a
in
e
R
e
c
ogni
s
e
d
41
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
19
T
r
a
in
e
R
e
c
ogni
s
e
d
42
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
20
T
r
a
in
e
R
e
c
ogni
s
e
d
43
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
21
T
r
a
in
e
R
e
c
ogni
s
e
d
44
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
22
T
r
a
in
e
R
e
c
ogni
s
e
d
45
T
e
s
t
N
ot
R
e
c
ogni
s
e
d
23
T
r
a
in
e
R
e
c
ogni
s
e
d
T
a
ble
2
.
R
e
s
ult
of
f
ace
r
e
c
ognit
ion
r
a
te
N
umbe
r
of
i
ma
ge
s
R
e
c
ogni
ti
on R
a
te
%
T
r
a
in
e
d i
ma
ge
30
100
T
e
s
te
d i
ma
g
e
15
86.667
T
ot
a
l
R
e
c
ogni
ti
on R
a
te
%
95.556
F
igur
e
11
.
Ac
c
ur
c
y
of
the
s
ys
tem
5.
CONC
L
USI
ON
I
n
thi
s
pa
pe
r
NF
ba
s
e
d
f
a
c
e
r
e
c
ognit
ion
s
ys
tem
h
a
s
be
e
n
int
r
oduc
e
d.
T
he
NN
s
ys
tem
c
ons
is
ted
of
two
-
s
tage
whic
h
a
r
e
tr
a
ini
ng
s
tage
a
nd
tes
ti
ng
s
t
a
ge
.
I
n
thi
s
pa
pe
r
,
two
s
e
ts
of
i
mage
a
r
e
us
e
d
to
e
va
luate
the
pe
r
f
or
manc
e
of
the
s
ys
tem.
One
of
the
s
e
t
is
us
e
d
in
the
tr
a
ini
ng
s
ta
ge
a
nd
whic
h
a
r
e
30
im
a
g
e
s
.
And
the
s
e
c
ond
s
e
t
of
im
a
ge
s
a
r
e
us
e
d
in
a
tes
ti
ng
s
tage
whic
h
is
15
im
a
ge
s
.
R
e
c
ognit
ion
r
a
te
us
e
d
in
thi
s
pa
pe
r
to
e
va
luate
the
pe
r
f
or
manc
e
o
f
the
s
ys
tem
a
nd
the
va
l
ue
of
r
e
c
ognit
ion
r
a
te
that
a
c
hieve
d
o
f
the
pr
opos
e
d
s
ys
tem
is
95.
556%
of
the
im
a
ge
s
a
r
e
oppos
e
d
r
e
c
og
nize
d
us
e
d
in
tr
a
ini
ng
a
nd
tes
ti
ng
s
tage
.
F
I
S
us
e
d
in
the
pa
pe
r
to
e
nha
nc
e
the
pe
r
f
or
manc
e
of
the
s
ys
tem
a
nd
to
ge
t
mor
e
a
c
c
ur
a
c
y
to
identi
f
y
the
r
e
s
ult
s
.
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
N
e
ur
o
-
fuz
z
y
inf
e
r
e
nc
e
s
y
s
tem
bas
e
d
face
r
e
c
ognit
i
on
us
ing
featur
e
e
x
tr
ac
ti
on
(
Ham
s
a
A
.
A
bdu
ll
ah)
435
RE
F
E
RE
NC
E
S
[1
]
Y
.
L
i
,
"
Face
Reco
g
n
i
t
i
o
n
Sy
s
t
em
,
"
San
g
w
h
a
n
Ch
a
Ph
D
T
h
e
s
i
s
,
2
0
1
9
.
[2
]
M.
A
.
H
amb
al
i
,
R.
G
.
J
i
mo
h
,
"
Perfo
rman
ce
E
v
al
u
a
t
i
o
n
o
f
Pri
n
ci
p
al
C
o
mp
o
n
e
n
t
A
n
al
y
s
i
s
A
n
d
I
n
d
e
p
en
d
en
t
Co
mp
o
n
e
n
t
A
n
a
l
y
s
i
s
A
l
g
o
ri
t
h
m
s
f
o
r
Faci
a
l
Reco
g
n
i
t
i
o
n
,
"
A
M
u
l
t
i
d
i
s
c
i
p
l
i
n
a
r
y
Jo
u
r
n
a
l
P
u
b
l
i
ca
t
i
o
n
o
f
t
h
e
F
a
cu
l
t
y
o
f
S
ci
e
n
ce,
,
v
o
l
.
1
2
,
p
p
.
4
7
-
6
2
,
2
0
1
5
.
[3
]
G
.
H
ap
s
ari
,
G
.
Mu
t
i
ara,
H
.
T
ari
g
a
n
,
"
Face
reco
g
n
i
t
i
o
n
s
mar
t
can
e
u
s
i
n
g
h
aar
-
l
i
k
e
feat
u
res
a
n
d
ei
g
en
fac
es
,
"
TE
LKO
M
NIK
A
Tel
ec
o
m
m
u
n
i
c
a
t
i
o
n
Co
m
p
u
t
i
n
g
E
l
ec
t
r
o
n
i
c
s
a
n
d
Co
n
t
r
o
l
,
v
o
l
.
1
7
,
n
o
.
2
,
p
p
.
9
7
3
-
9
8
0
,
2
0
1
9
.
[4
]
H
.
Pras
et
y
o
,
B.
A
k
a
rd
i
h
a
s
,
"
Bat
i
k
i
mag
e
re
t
ri
e
v
al
u
s
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
ral
n
et
w
o
r
k
,
"
TE
LKO
M
NI
KA
Tel
eco
m
m
u
n
i
ca
t
i
o
n
Co
m
p
u
t
i
n
g
E
l
ect
r
o
n
i
c
s
a
n
d
Co
n
t
r
o
l
,
v
o
l
.
1
7
,
n
o
.
6
,
p
p
.
3
0
1
0
-
3
0
1
8
,
2
0
1
9
.
[5
]
Su
t
i
k
n
o
,
H
.
A
.
W
i
b
a
w
a,
P.
S.
Sas
o
n
g
k
o
,
"
D
et
ect
i
o
n
o
f
Sh
i
p
u
s
i
n
g
Imag
e
Pr
o
ces
s
i
n
g
an
d
N
e
u
ral
N
et
w
o
r
k
,
"
TE
LKO
M
NIK
A
Tel
ec
o
m
m
u
n
i
c
a
t
i
o
n
Co
m
p
u
t
i
n
g
E
l
ec
t
r
o
n
i
c
s
a
n
d
Co
n
t
r
o
l
,
v
o
l
.
1
6
,
n
o
.
1
,
p
p
.
2
5
9
-
2
6
4
,
2
0
1
8
.
[6
]
O
.
A
L
-
A
l
l
af,
A
.
T
ami
m
i
,
M.
A
l
i
a,
"
Face
Reco
g
n
i
t
i
o
n
Sy
s
t
em
Bas
e
d
o
n
D
i
ff
eren
t
A
rt
i
fi
c
i
al
N
e
u
ral
N
e
t
w
o
rk
s
Mo
d
el
s
an
d
T
ra
i
n
i
n
g
A
l
g
o
r
i
t
h
ms
,
"
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
A
d
va
n
ce
d
Co
m
p
u
t
er
S
c
i
en
ce
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
4
,
n
o
.
6
,
p
p
.
4
0
-
4
7
,
2
0
1
3
.
[7
]
V
.
Bh
at
an
d
J
.
Pu
j
ari
,
"
N
eu
r
o
-
fu
zz
y
fu
s
i
o
n
i
n
a
mu
l
t
i
mo
d
a
l
face
reco
g
n
i
t
i
o
n
u
s
i
n
g
PCA
,
ICA
an
d
SIFT
,
"
In
t
.
J.
Co
m
p
u
t
a
t
i
o
n
a
l
V
i
s
i
o
n
a
n
d
R
o
b
o
t
i
cs
,
,
v
o
l
.
4
,
n
o
.
6
,
p
p
.
4
1
4
-
4
3
4
,
2
0
1
6
.
[8
]
S.
H
amd
an
,
A
.
Sh
ao
u
t
,
"
Face
Reco
g
n
i
t
i
o
n
U
s
i
n
g
N
eu
r
o
-
Fu
zzy
A
n
d
E
i
g
en
face,
"
In
t
er
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
Co
m
p
u
t
e
r
S
ci
e
n
ce
a
n
d
E
n
g
i
n
ee
r
i
n
g
(IJCS
E
),
v
o
l
.
5
,
n
o
.
4
,
p
p
.
1
-
1
0
,
2
0
1
6
.
[9
]
T
.
Ch
an
d
ras
e
k
h
ar
,
C.
k
u
mar,
"
Face
Reco
g
n
i
t
i
o
n
S
y
s
t
em
u
s
i
n
g
A
d
a
p
t
i
v
e
N
eu
r
o
f
u
zz
y
In
fere
n
ce
Sy
s
t
e
m,
"
in
In
t
er
n
a
t
i
o
n
a
l
Co
n
f
e
r
en
ce
o
n
E
l
ec
t
r
i
ca
l
,
E
l
ec
t
r
o
n
i
cs
,
C
o
m
m
u
n
i
ca
t
i
o
n
,
Co
m
p
u
t
er
a
n
d
O
p
t
i
m
i
z
a
t
i
o
n
Tech
n
i
q
u
es
(ICE
E
CC
O
T),
IE
E
E
,
My
s
u
ru
,
In
d
i
a,
2
0
1
7
.
[1
0
]
J
.
L
i
,
T
.
Q
i
u
,
C.
W
en
,
K
.
X
i
e,
F.
W
en
,
"
Ro
b
u
s
t
Face
Reco
g
n
i
t
i
o
n
U
s
i
n
g
t
h
e
D
ee
p
C2
D
-
C
N
N
M
o
d
e
l
Bas
ed
o
n
D
eci
s
i
o
n
-
L
ev
e
l
Fu
s
i
o
n
,
"
S
en
s
o
r
s
,
v
o
l
.
1
8
,
n
o
.
7
,
p
p
.
1
-
2
7
,
2
0
1
8
.
[1
1
]
V
.
G
o
s
av
i
,
A
.
D
e
s
h
ma
n
e,
G
.
Sab
l
e,
"
A
d
ap
t
i
v
e
N
e
u
ro
F
u
zzy
In
fere
n
ce
Sy
s
t
em
f
o
r
Faci
a
l
Reco
g
n
i
t
i
o
n
,
"
IO
S
R
J
o
u
r
n
a
l
o
f
E
l
ect
r
i
c
a
l
a
n
d
E
l
ec
t
r
o
n
i
cs
E
n
g
i
n
eer
i
n
g
,
v
o
l
.
1
4
,
n
o
.
3
,
p
p
.
1
5
-
2
2
,
2
0
1
9
.
[1
2
]
M.
N
as
ru
d
i
n
,
S.
Y
aak
o
b
,
I.
I
s
zai
d
y
,
A
.
A
b
d
u
l
-
N
a
s
i
r,
"
Imag
e
E
x
t
ract
i
o
n
u
s
i
n
g
G
eo
me
t
ri
c
an
d
Z
ern
i
k
e
Mo
m
en
t
In
v
ar
i
an
t
s
,
"
i
n
In
t
er
n
a
t
i
o
n
a
l
P
o
s
t
g
r
a
d
u
a
t
e
Co
n
f
er
e
n
ce
o
n
E
n
g
i
n
eer
i
n
g
a
n
d
M
a
n
a
g
e
m
en
t
,
Mal
ay
s
i
a,
2
0
1
4
.
[1
3
]
V
J
ai
s
w
al
,
V
.
Sh
arma,
S.
V
arma,
"
A
n
i
mp
l
emen
t
at
i
o
n
o
f
n
o
v
e
l
g
en
e
t
i
c
b
a
s
ed
cl
u
s
t
er
i
n
g
al
g
o
ri
t
h
m
fo
r
co
l
o
r
i
mag
e
s
eg
me
n
t
a
t
i
o
n
,
"
TE
LKO
M
NIK
A
Tel
eco
m
m
u
n
i
c
a
t
i
o
n
Co
m
p
u
t
i
n
g
E
l
ec
t
r
o
n
i
cs
a
n
d
Co
n
t
r
o
l
,
v
o
l
.
1
7
,
n
o
.
3
,
p
p
.
1
4
6
1
-
1
4
6
7
,
2
0
1
9
.
[1
4
]
L
.
K
o
t
o
u
l
a
s
,
I.
A
n
d
r
ea
d
i
s
,
"
Imag
e
A
n
a
l
y
s
i
s
U
s
i
n
g
Mo
m
en
t
s
,
"
5
t
h
In
t
.
Co
n
f
.
o
n
Tech
n
o
l
o
g
y
a
n
d
A
u
t
o
m
a
t
i
o
n
,
,
2
0
0
5
.
[1
5
]
R.
K
ap
o
o
r,
P.
Mat
h
u
r,
"
Face
Reco
g
n
i
t
i
o
n
U
s
i
n
g
Mo
m
en
t
s
an
d
W
a
v
el
e
t
s
,
"
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
E
n
g
i
n
ee
r
i
n
g
R
es
e
a
r
c
h
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
3
,
n
o
.
4
,
p
p
.
8
2
-
9
5
,
2
0
1
3
.
[1
6
]
M.
A
b
d
a
l
a,
B.
K
h
amma
s
,
H
.
A
b
d
u
l
l
ah
,
"
E
y
e
-
Id
e
n
t
i
f
i
c
at
i
o
n
S
y
s
t
em
Bas
e
d
o
n
Back
-
Pr
o
p
a
g
at
i
o
n
N
N
Cl
a
s
s
i
fi
er,
"
Jo
u
r
n
a
l
o
f
E
n
g
i
n
ee
r
i
n
g
a
n
d
D
evel
o
p
m
en
t
,
v
o
l
.
1
4
,
n
o
.
4
,
p
p
.
3
4
-
5
0
,
2
0
1
0
.
[1
7
]
S.
Si
l
ak
ar
i
,
M.
Mo
t
w
a
n
i
,
M.
Mah
e
s
h
w
ari
,
"
Co
l
o
r
Imag
e
Cl
u
s
t
er
i
n
g
u
s
i
n
g
B
l
o
c
k
T
r
u
n
ca
t
i
o
n
A
l
g
o
ri
t
h
m,
"
In
t
er
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
Co
m
p
u
t
e
r
S
ci
e
n
ce
Is
s
u
e
s
,
v
o
l
.
4
,
n
o
.
2
,
p
p
.
3
1
-
3
5
,
2
0
0
9
.
[1
8
]
A
fi
f
i
an
d
W
.
A
s
h
o
u
r,
"
Imag
e
Re
t
ri
e
v
al
Ba
s
e
d
o
n
Co
n
t
e
n
t
U
s
i
n
g
Co
l
o
r
Fea
t
u
re,
"
In
t
er
n
a
t
i
o
n
a
l
S
ch
o
l
a
r
l
y
R
e
s
ea
r
ch
Net
wo
r
k,
v
o
l
.
2
0
1
2
,
p
p
. 1
-
1
2
,
2
0
1
2
.
[1
9
]
Su
g
i
a
rt
i
,
Y
.
Y
u
h
a
n
d
r,
J
.
N
aam,
D
.
In
d
ra,
J
.
Sa
n
t
o
n
y
,
"
A
n
art
i
f
i
ci
a
l
n
e
u
ral
n
e
t
w
o
rk
a
p
p
r
o
ach
fo
r
d
et
ec
t
i
n
g
s
k
i
n
can
c
er,
"
TE
LKO
M
NIK
A
Tel
ec
o
m
m
u
n
i
c
a
t
i
o
n
Co
m
p
u
t
i
n
g
E
l
ec
t
r
o
n
i
c
s
a
n
d
Co
n
t
r
o
l
,
v
o
l
.
1
7
,
n
o
.
2
,
p
p
.
7
8
8
-
7
9
3
,
2
0
1
9
.
[2
0
]
S.
Meh
t
a,
S.
G
u
p
t
a,
B.
Bh
u
s
h
an
a
n
d
C.
K
.
N
ag
p
a
l
,
"
Face
Reco
g
n
i
t
i
o
n
u
s
i
n
g
N
eu
r
o
-
Fu
zzy
In
fere
n
ce
Sy
s
t
e
m,
"
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
S
i
g
n
a
l
P
r
o
ce
s
s
i
n
g
,
Im
a
g
e
P
r
o
c
es
s
i
n
g
a
n
d
P
a
t
t
er
n
R
eco
g
n
i
t
i
o
n
,
v
o
l
.
7
,
n
o
.
1
,
p
p
.
3
3
1
-
3
4
4
,
2
0
1
4
.
[2
1
]
K
.
Sarav
an
an
,
S.
Sas
i
t
h
r
a,
"
Rev
i
ew
o
n
Cl
a
s
s
i
fi
ca
t
i
o
n
Ba
s
ed
o
n
A
r
t
i
f
i
ci
a
l
N
eu
ra
l
N
et
w
o
r
k
s
,
"
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
A
m
b
i
e
n
t
S
y
s
t
e
m
s
a
n
d
A
p
p
l
i
ca
t
i
o
n
s
(IJA
S
A
),
v
o
l
.
2
,
n
o
.
4
,
p
p
.
1
1
-
1
8
,
2
0
1
4
.
[2
2
]
R.
V
y
as
,
G
.
G
arg
,
"
F
ace
reco
g
n
i
t
i
o
n
u
s
i
n
g
feat
u
re
ex
t
ra
ct
i
o
n
an
d
n
eu
r
o
-
fu
zz
y
t
ec
h
n
i
q
u
es
,
"
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
E
l
ec
t
r
o
n
i
cs
a
n
d
C
o
m
p
u
t
er
S
c
i
en
ce
E
n
g
i
n
eer
i
n
g
,
v
o
l
.
1
,
n
o
.
4
,
p
.
1
0
,
2
0
1
2
.
[2
3
]
Bi
s
t
,
"
Fu
zz
y
L
o
g
i
c
fo
r
C
o
mp
u
t
er
V
i
ru
s
D
et
e
ct
i
o
n
,
"
In
t
e
r
n
a
t
i
o
n
a
l
Jo
u
r
n
a
l
o
f
E
n
g
i
n
ee
r
i
n
g
S
c
i
en
ce
s
&
R
es
a
r
ch
Tech
n
o
l
o
g
y,
v
o
l
.
3
,
n
o
.
2
,
p
p
.
7
71
-
7
7
3
,
2
0
1
4
.
[2
4
]
T
.
T
u
n
cer,
S.
D
o
g
an
,
M.
A
b
d
ar,
M.
Bas
i
r
i
,
P.
Pl
a
w
i
a
k
,
"
Face
Reco
g
n
i
t
i
o
n
w
i
t
h
T
ri
a
n
g
u
l
ar
Fu
zz
y
Set
-
B
as
e
d
L
o
cal
Creo
o
Pat
t
eren
i
n
W
a
v
el
e
t
D
o
mai
n
,
"
S
ym
m
e
t
r
y,
v
o
l
.
1
1
,
n
o
.
6
,
p
p
.
1
-
1
8
,
2
0
1
9
.
[2
5
]
K
.
Bah
rei
n
i
,
W
.
V
e
g
t
,
W
.
W
e
s
t
era,
"
A
F
u
zzy
L
o
g
i
c
A
p
p
ro
ac
h
t
o
Rel
i
a
b
l
e
Rea
l
-
T
i
me
Reco
g
n
i
t
i
o
n
o
f
Fa
ci
al
E
mo
t
i
o
n
s
,
"
M
u
l
t
i
m
ed
i
a
To
o
l
s
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
7
8
,
n
o
.
1
4
,
p
p
.
1
8
9
4
3
-
1
8
9
6
6
,
2
0
1
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.