Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
23
,
No.
3
,
Septem
ber
20
21
,
pp.
1
847
~
1854
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v
2
3
.i
3
.
pp
1847
-
1854
1847
Journ
al h
om
e
page
:
http:
//
ij
eecs.i
aesc
or
e.c
om
Identific
atio
n
of
optimu
m segme
nt in sin
gle ch
annel
EEG
biomet
ric s
ystem
Muhamm
ad
Af
if
Hen
draw
an
,
Pr
amana
Yog
a
S
aputr
a, C
ahy
a Rahm
ad
Depa
rtment
o
f
I
nform
at
ion
T
ec
h
nolog
y
,
Pol
it
ekn
ik
Nege
r
i
Ma
la
n
g,
Indone
si
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ma
y
21
,
2021
Re
vised
J
ul
30
2021
Accepte
d
Aug
6
,
2021
Now
aday
s,
bio
m
et
ric
m
odalitie
s
have
g
ai
ned
p
opula
rity
in
s
ec
u
rity
s
y
s
te
m
s.
Neve
rthele
ss
,
th
e
conve
nt
iona
l
c
om
m
erc
ia
l
-
gra
d
e
biometri
c
s
y
s
tem
addr
esses
som
e
issues.
The
bigge
st
probl
e
m
is
tha
t
they
c
an
be
impos
ed
b
y
ar
ti
fi
cial
biometri
cs.
The
el
e
ct
ro
encepha
l
ogra
m
(EE
G)
i
s
a
poss
ibl
e
sol
uti
on.
I
t
i
s
nea
rl
y
impos
sible
to
rep
licate
bec
ause
i
t
is
depe
nden
t
on
hum
an
m
ent
al
ac
t
ivi
t
y
.
Sever
al
studie
s
hav
e
al
r
ea
d
y
d
emons
tra
t
ed
a
h
igh
l
evel
of
accurac
y
.
How
eve
r,
it
r
equ
ire
s
a
l
arg
e
num
ber
of
s
ensors
a
nd
ti
m
e
to
col
l
ect
th
e
signa
l.
Thi
s
stud
y
prop
osed
a
biometr
ic
s
y
stem
using
si
ngle
-
ch
anne
l
EEG
rec
orde
d
during
rest
ing
e
y
es
o
pen
(
EO)
c
ondit
ions.
A
total
of
45
EE
G
si
gnal
s
from
9
subjec
ts
wer
e
c
oll
e
ct
ed
.
Th
e
E
EG
signal
was
segm
ent
ed
in
t
o
5
sec
ond
le
ngths.
The
a
lp
ha
band
was
used
in
thi
s
stud
y
.
Discre
t
e
wav
elet
t
ran
sform
(DW
T)
with
Daube
chies
t
y
p
e
4
(db4)
was
emp
lo
y
ed
to
ext
r
act
the
al
pha
band.
Pow
er
spec
tral
density
(PS
D)
was
ext
rac
t
e
d
from
ea
ch
segm
ent
as
the
m
ai
n
fea
ture.
Li
ne
ar
discri
m
i
nant
ana
l
y
sis
(
LDA)
and
sup
port
vec
tor
m
ac
hine
(SV
M)
were
used
to
cla
ss
if
y
the
EE
G
signal
.
The
propo
sed
m
et
hod
ac
hi
eve
d
86%
a
cc
ura
c
y
using
L
DA
onl
y
from
t
he
t
hird
s
egment.
Th
ere
for
e
,
thi
s
stud
y
show
ed
that
it
is
poss
ibl
e
to
ut
il
i
ze
s
ingl
e
-
cha
nne
l
E
EG
during
a
resti
ng
EO
sta
te
in
a
biometr
ic
s
ystem.
Ke
yw
or
ds:
Bi
om
e
tric
EEG
LDA
SV
M
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Muh
am
m
ad
Afi
f
He
ndra
wan
Dep
a
rtm
ent o
f Info
rm
at
ion
Te
chnolo
gy
Po
li
te
kn
i
k Nege
ri Mal
ang
So
e
karn
o
-
Hatt
a St.
No.
9
Ma
l
ang Ci
ty
, East
Java,
Indo
nesi
a
Em
a
il
:
afif.h
en
dr
a
wa
n@gm
ai
l
.co
m
1.
INTROD
U
CTION
On
e
of
the
crit
ic
al
factor
s
in
asset
pr
otect
io
n
is
a
secur
it
y
syst
e
m
.
The
secur
it
y
syst
e
m
s
hould
m
eet
sever
al
requir
e
m
ents,
su
c
h
as
diff
ic
ult
to
pe
netrate,
i
m
ple
m
enting
the
a
uth
e
ntica
ti
on
m
et
ho
d,
a
nd
auth
or
iz
at
io
n
to
pro
vid
e
a
s
ecur
e
sec
uri
ty
syst
e
m
.
The
a
uth
e
ntica
ti
on
m
et
ho
d
is
an
essenti
al
par
t
of
th
e
secur
it
y
syst
em
.
It
ver
ifie
s
the
us
e
r
be
f
or
e
bein
g
able
to
a
ccess
the
asset
s.
The
c
om
bin
at
ion
of
use
r
na
m
e
and
pass
word
is
t
he
m
os
t
widely
adopted
a
uth
e
nt
ic
at
ion
m
et
hod.
Howe
ver
,
it
has
bee
n
pro
ve
d
that
the
u
ser
nam
e
and
pass
w
ord
had
m
any
vulnera
bili
ti
es.
Their
c
om
bin
at
i
on
is
easy
t
o
f
orget,
a
nd
it
c
an
be
stolen
usi
ng
a
dicti
on
a
ry
brut
e
force
at
ta
ck
.
A
no
t
her
a
ppr
oach
is
the
c
oncept
of
“s
ome
thing
yo
u
ha
ve”
[
1]
.
It
m
e
ans
the
pro
of
of
ide
ntit
y
is
rep
resen
t
ed
by
an
obj
ec
t
wh
ic
h
the
use
r
has.
It
can
be
an
identit
y
card,
sm
art
card
,
rad
i
o
fr
e
qu
e
ncy
ide
nt
ific
at
ion
de
vic
e
(
RFI
D
)
,
a
nd
so
on.
T
her
e
a
re
so
m
e
issues
accor
ding
t
o
the
“s
om
et
hin
g
yo
u
hav
e”
conce
pt.
Th
e
ob
j
ec
t can
b
e
stolen
, dam
aged, a
nd it
is
no
t
flexi
ble.
Du
e
t
o
ad
va
nc
es
in
te
chnolo
gy
,
it
is
po
ssible
to
colle
ct
hu
m
an
trai
ts
as
a
bi
om
et
ric
m
od
al
i
ty
[2]
,
[3]
.
Re
searche
rs
tr
ie
d
to
stu
dy
the
po
s
sibil
it
ie
s
of
fi
ng
e
r
pr
i
nt,
vo
ic
e,
face
,
iris,
a
nd
palm
pr
int
in
bio
m
et
ric
auth
e
ntica
ti
on
syst
e
m
s
[2]
-
[6]
.
Fing
e
r
p
ri
nt
m
od
al
it
y
is
the
m
os
t
co
m
m
on
com
m
ercial
gr
ade
bi
om
et
ric
syst
e
m
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
2
3
, N
o.
3
,
Se
ptem
ber
20
21
:
18
47
-
18
54
1848
in
the
m
ark
et
.
It
is
easy
to
colle
ct
the
dat
a
an
d
re
quires
low
com
pu
ta
ti
on
al
c
os
t
in
pract
ic
e.
N
onet
heless
,
fin
gerpr
i
nts
fa
ce
so
m
e
issues
in
sec
ur
it
y.
O
ne
of
w
hich
is
it
is
possible
to
re
pl
ic
at
e
a
r
eal
fin
ger
us
i
ng
a
n
arti
fici
al
f
ing
e
r
[
7]
.
Mor
e
ov
e
r,
a
nother
m
ark
et
-
re
ady
bio
m
et
ric
authen
ti
cat
io
n
syst
e
m
is
a
lso
facin
g
the
s
a
m
e
issue.
Vo
ic
e
m
od
al
it
y
can
be
i
m
i
ta
t
ed
us
i
ng
play
ba
ck
spo
of
in
g
[
5]
.
Face,
iris,
a
nd
palm
pr
int
im
age
-
base
d
tra
it
s
can
be
rep
li
cat
ed
usi
ng
a
high
-
res
olu
ti
on
i
m
age
[8]
.
The
m
ai
n
reason
these
m
od
al
it
ie
s
can
be
i
m
it
at
ed
is
becau
se
they
are
e
xpose
d.
T
he
refor
e
,
to
pro
vid
e
a
hi
gh
-
sec
ur
it
y
biom
et
ric
authen
ti
cat
ion
syst
em
,
inv
isi
ble
m
od
a
li
ti
es
that are
not ex
po
s
ed
fro
m
o
ut
side are
n
ee
de
d.
Hu
m
an
bi
os
ig
nal
is
a
po
te
ntial
m
od
al
it
y
to
reso
l
ve
that
iss
ue.
It
is
cu
rr
e
nt
ly
gaining
po
pu
la
rity
as
a
m
od
al
ity
fo
r
nonm
edical
us
e
[9]
-
[12]
.
Bi
osi
gn
al
is
a
n
el
ect
rical
sign
al
pro
du
ce
d
by
hum
an
body
ac
ti
vity
.
Ele
ct
ro
e
nceph
al
ogram
(EEG)
is
an
exam
p
le
of
a
bio
sig
nal.
It
captu
re
s
the
el
ect
rica
l
sign
al
fr
om
br
ai
n
act
ivit
y.
Ma
r
cel
and
Mi
ll
an
[
13
]
fou
nd
t
hat
the
EE
G
sig
na
l
is
un
i
qu
e
am
ong
t
he
oth
er
s.
The
refor
e
,
it
c
an
be
us
e
d
as a b
iom
et
ric authe
ntica
ti
on
m
od
al
it
y.
EEG has sev
e
r
al
ad
va
ntages
c
om
par
ed
to c
onve
ntio
nal m
o
dalit
y.
First,
EE
G
is
inv
isi
ble,
a
nd
it
can
not
be
captu
red
us
i
ng
rem
ote
sensing
[
14]
.
Sec
ond,
th
e
EE
G
si
gn
al
is
dep
e
ndent
on
the
m
ental
con
diti
on
of
an
ind
ivi
du
al
[15]
.
A
m
ental
c
onditi
on
s
uc
h
as
stress,
m
oo
d,
a
nd
pr
ess
ure
will
aff
ect
the
EE
G
sign
al
.
Thi
rd,
the
EE
G
si
gn
al
pro
ved
that
t
he
sig
nal
is
c
om
ing
f
ro
m
a
li
ving
ind
ivi
du
al
[
1]
.
Seve
ral
stu
die
s
ha
ve
bee
n
propose
d
t
o
us
e
EEG
a
s
a
biom
et
ric
authen
t
ic
at
ion
m
od
al
it
y.
It
can
be
div
ide
d
i
nto
th
ree
m
ai
n
ca
te
go
ries
base
d
on
their
sti
m
uli.
The
first
cat
eg
or
y
is
based
on
resti
ng
sta
te
,
th
e
seco
nd
cat
eg
ory
is
visu
al
sti
m
ul
i,
and
the
l
ast
is
base
d
on
m
oto
r
stim
uli.
Para
njape
et
al.
[
16]
is
a
nota
ble
stud
y
in
t
he
e
arly
sta
ge
of
t
his
stu
dy.
T
he
y
us
e
ei
ght
-
c
ha
nn
el
E
EG
at
F7
,
F8,
T
3,
T
4,
T
5,
T6
,
P
3,
an
d
P
4
po
sit
io
ns
acc
or
ding
to
the
10
-
20
syst
em
.
Au
tore
gr
e
ssive
(AR
)
m
od
el
featu
res
a
nd
disc
rim
inan
t
analy
sis
we
r
e
e
m
plo
ye
d
to
cl
assify
40
sub
j
e
ct
s
based
on
re
sti
ng
-
sta
te
in
ey
e
-
openi
ng
a
nd
ey
e
-
cl
os
e
d
c
onditi
ons.
T
he
stud
y
achieve
d
80
%
accuracy
by
usi
ng
a
50%
data
sam
ple
as
train
in
g
data
an
d
100%
acc
urac
y
by
us
in
g
al
l
data
as
trai
ning
data.
Howe
ver,
[
16
]
e
m
plo
ys
a
high
nu
m
ber
of
e
poch
an
d
high
or
der
of
AR
to
a
chieve
that
acc
ur
acy
.
The
ef
fect
of
t
he
resti
ng
-
sta
te
al
so
was
n’t
cl
early
r
ep
or
te
d.
Pala
niap
pa
n
and
Ma
ndic
[
17
]
propose
d
a
visu
al
stim
uli
m
et
ho
d
du
ri
ng
the
E
E
G
recor
ding
session.
Sub
j
ect
s
are
require
d
to
rem
e
m
ber
a
picture
that
shows
in
the
visu
al
sti
m
uli.
61
EE
G
c
hannel
was
uti
li
zed
in
[
17
]
,
and
it
gai
ns
98.
12%
acc
ur
ac
y
by
us
in
g
e
xt
end
e
d
near
est
nei
ghbor
(E
N
N)
.
T
he
m
ulti
ple
sign
al
cl
assifi
cat
ion
(M
US
IC
)
a
lgorit
hm
was
e
m
plo
ye
d
i
n
[
17
]
t
o
extract EE
G fe
at
ur
es,
but
on
l
y t
he
po
wer sp
ect
ru
m
co
m
po
nen
t
was
u
se
d.
The
EE
G
m
oto
r
im
aginar
y
(
EEGMM
I
DB)
dataset
f
ro
m
Ph
ysi
oNet
[18]
has
bec
om
e
popula
r
in
bio
m
et
ric
stud
i
es.
T
he
data
w
as
init
ia
ll
y
des
ign
e
d
for
a
br
a
in
-
c
om
pu
te
r
in
te
rf
ace
(BCI)
s
tud
y.
EE
GMMIDB
consi
sts
of
4
t
asks
of
m
oto
r
i
m
aginar
y,
an
d
two
baseli
ne
conditi
ons.
Re
sti
ng
-
sta
te
with
ey
e
op
e
n
(E
O)
a
nd
ey
e
cl
os
e
(
EC)
are
the
basel
ine
co
ndit
ion
s
in
EE
GMMI
DB
[
18]
.
O
nly
resti
ng
E
O
a
nd
resti
ng
EC
widel
y
repor
te
d
we
re u
se
d
in the b
i
om
et
ric stud
y
[19]
-
[
22]
. Ro
cca
et
a
l.
[19]
e
m
plo
ye
d
EO
a
nd
EC baseli
ne
from
the
EEGMM
I
DB
dataset
.
T
hey
pro
po
se
d
sp
ec
tral
co
her
e
nce
co
nn
e
ct
ivit
y,
a
m
ulti
-
br
ai
n
reg
i
on
f
us
e
spe
ct
ral
coh
e
re
nce.
Th
ei
r
stud
y
sho
wed
th
at
co
nn
ect
ivit
y
between
brai
n
re
gions
im
pr
ov
es
EEG
-
base
d
bi
om
et
ric
p
e
r
f
o
r
m
a
n
c
e
r
a
t
h
e
r
t
h
a
n
p
o
w
e
r
s
p
e
c
t
r
um
e
s
t
i
m
a
t
i
o
n
f
r
o
m
a
s
i
n
g
l
e
b
r
a
i
n
r
e
g
i
o
n
.
M
o
r
e
o
v
e
r
,
F
r
a
s
c
h
i
n
i
e
t
a
l
.
[20]
con
ti
nue
d
stu
dy
ing
brai
n
co
nnect
ivit
y
base
d
on
[19]
by
pro
po
si
ng
ei
ge
nv
ect
or
ce
ntral
it
y
instea
d
of
whole
functi
onal
brai
n
c
onnecti
vity
.
They
a
ppli
ed
Eucli
dea
n
distance
to
c
om
par
e
eac
h
f
eat
ur
e
set
,
a
nd
they
us
e
d
false
acce
ptanc
e
rate
(FAR)
a
nd
false
re
j
ect
ion
rate
(F
RR
)
as
an
eval
uato
r
.
F
rasch
i
ni
et
al
.
[
20
]
repo
rted
that
the
eq
ual
er
ror
rate
(EER
)
f
r
om
the
gam
ma
band
is
0.044,
a
nd
EER
f
r
om
hig
h
beta
is
0.102.
It
wa
s
al
so
repor
te
d
that
a
low
beta
band
s
howe
d
a
lo
wer
rate
wit
h
0.144
of
EER.
Th
om
as
and
Vinod
[
21
]
propos
e
d
non
-
interc
onne
ct
ivit
y
br
ai
n
f
un
ct
io
n
i
n
t
he
ir
stu
dy
f
ro
m
the
sam
e
dataset
.
They
em
pl
oy
sam
ple
entropy
(S
am
pEn
)
fro
m
5
band
sig
na
ls,
delta
,
theta
,
al
ph
a
,
beta,
a
nd
gam
m
a.
Their
stud
y
re
por
te
d
an
av
era
ge
correct
recog
niti
on
rat
e
of
98.31%
.
They
al
s
o
re
porte
d
the
e
xtra
ct
ed
power
s
pe
ct
ral
de
ns
it
y
(P
S
D)
f
ro
m
Sa
m
pEn
enh
a
nce
d
co
rrec
ti
on
rate
up
to
99.
7%.
Kang
et
al.
[
22
]
tr
ie
d
to
re
du
ce
t
he
num
ber
of
channels.
T
he
fron
ta
l
c
hannel
c
on
ta
i
ns
ey
e
m
ov
em
ent
act
ivit
y
that
ge
ner
at
es
an
el
ect
rooc
ulogram
(EOG)
si
gn
al
dur
ing
E
O
sta
te
[15]
,
[22]
,
[
23
]
.
Mo
r
eov
e
r,
t
he
oc
ci
pital
chan
ne
l
gen
er
at
ed
a
burst
of
hi
gh
al
pha
sig
na
ls
that
inco
ns
ist
ently
app
ea
r
duri
ng
EC
sta
te
[22]
.
Ther
e
f
or
e,
the
y
exclud
e
d
22
channels
f
ro
m
fron
ta
l
an
d
oc
ci
pital
reg
i
on
s
due
to
no
ise
iss
ues.
As
a
fi
nal
res
ul
t,
Kang
et
al.
[22]
only
e
m
plo
y
34
cha
nnel
s
in
their
stu
dy
.
Ten
sing
le
-
c
ha
nn
el
and
te
n
m
ulti
-
channel
feat
ures,
inclu
ding
P
SD
,
wer
e
e
xtra
ct
ed.
T
hey
repor
te
d
0.7
3%
of
EER
and 1.8
0%
of
EER d
ur
i
ng E
O
a
nd EC,
res
pe
ct
ively
.
The
previ
ous
stud
ie
s
s
howe
d
outst
an
ding
resu
lt
s
by
us
i
ng
m
ulti
ple
c
hannels.
Howe
ve
r
,
a
fewe
r
nu
m
ber
of
c
ha
nn
el
s
or
eve
n
s
ing
le
-
c
ha
nn
el
EEG
is
m
or
e
r
easo
nab
le
i
n
pract
ic
e.
The
stud
y
of
sin
gle
-
c
hann
e
l
EEG
is
ra
rely
r
eported
.
S
uppi
ah
an
d
Vinod
[
15
]
colle
ct
ed
t
he
EE
G
sig
nal
from
the
O2
ch
ann
el
d
uri
ng
th
e
EC
sta
te
.
They
repor
te
d
that
t
he
al
ph
a
c
ha
nnel
is
an
e
ff
ect
ive
ba
nd
com
bin
e
d
with
P
SD
fe
at
ur
es
t
o
disti
nguis
h
on
e
sel
f
from
oth
ers.
Fis
her’s
li
near
discrim
inant
analy
sis
(F
L
DA)
was
e
m
plo
ye
d
to
exa
m
ine
the
pr
opose
d
m
et
ho
d.
Howe
ver,
Sup
piah
a
nd
Vino
d
[15],
repor
te
d
t
hat
the
pro
posed
m
et
hod
re
qu
i
res
m
or
e
than
5
se
conds
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c
En
g
&
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Id
e
ntif
ic
ation
of
o
ptim
um se
gment i
n
si
ng
le
chan
nel EE
G b
iometric
syste
m
(
Mu
ham
m
ad Afi
f Hendr
aw
an
)
1849
of
EE
G
si
gn
al
to
ac
hieve
97
-
99%
of
accu
ra
cy
.
Zey
nali
a
nd
Seye
dar
a
bi
[
24
]
al
s
o
re
ported
t
hat
the
O
2
channel
is
the
opti
m
u
m
cha
nn
el
place
m
ent
for
bio
m
et
ric
EEG
.
T
he
y
reach
95%
a
ccur
acy
i
n
a
la
rg
e
dataset
on
l
y
with
the
O2
c
hanne
l.
Desp
it
e
that,
in
[24]
,
the
E
EG
sig
nal
was
c
ollec
te
d
with
five
m
ental
a
ct
ivit
ie
s.
This
stud
y
addresses
s
om
e
issues
m
entione
d
a
bove
,
especial
ly
in
data
colle
ct
ion
proce
du
re
an
d
proce
ssing
i
n
sing
le
-
c
ha
nn
el
EEG
as
a
biom
et
ric
m
od
al
ity.
This
st
ud
y
s
hows
t
hat
it
is
possible
t
o
e
m
plo
y
sing
le
-
c
hanne
l
EEG
a
s a
bio
m
et
ric m
od
al
it
y
durin
g resti
ng
EO
c
onditi
ons.
2.
RESEA
R
CH MET
HO
D
2.1.
Data
c
ollec
tion
The
r
aw
E
EG
sign
al
s
wer
e
colle
ct
ed
us
i
ng
the
Ne
uros
ky
Mi
nd
wa
ve
Mob
il
e
hea
ds
e
t.
It
us
es
a
dry
-
ty
pe
sin
gl
e
-
cha
nnel
se
nsor
place
d
at
Fp1
acc
ordin
g
to
t
he
10
-
20
placem
ent
syst
e
m
.
The
refe
ren
c
e
el
ect
ro
de
was
placed
in
the
le
ft
ear.
It
was
placed
in
the
A1
po
sit
io
n
an
d
a
512
Hz
sa
m
pl
ing
rate
wa
s
us
ed
.
Nine
par
ti
ci
pa
nts
pa
rtic
ipate
d
in
t
his
resea
r
ch
(
8
m
al
es,
1
fem
al
e,
M
=
20
.
66).
All
pa
rtic
ipants
had
no
rm
al
or
correct
ed
to
norm
al
vision
,
norm
al
aud
it
ion
,
an
d
ri
gh
t
-
ha
nd
e
d.
All
par
ti
ci
pan
ts
al
so
ne
ver
e
xp
e
rience
d
any
chro
nic
diseas
e.
Co
ns
ide
rin
g
the
neur
os
ky
m
ind
wa
ve
m
ob
il
e
hea
ds
et
use
s
a
dry
el
ect
r
od
e
,
it
does
n’
t
requir
e
sp
eci
al
pre
par
a
ti
on
.
It
only
to
ok
le
ss
t
han
10
sec
onds
to
prepa
re
t
he
e
quipm
ent.
Partic
ipants
we
re
a
ske
d
t
o
blink
natural
ly
du
ri
ng
data
acqu
isi
ti
on.
Eac
h
par
ti
ci
pa
nt
ha
s
per
f
orm
ed
five
tria
l
data
c
ollec
ti
on
proce
dures
with
60
seco
nd
s
e
ach
tria
l.
30s
second
resti
ng
tim
e
was
giv
en
to
pa
rtic
ipants
was
gi
ven
betwee
n
each
t
rial
.
A
total
o
f
45 raw
EEG si
gnal
s were
ob
ta
ine
d
i
n t
he data
ac
qu
i
sit
ion
procedu
r
e.
2.2.
Metho
dol
ogy
T
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
i
s
d
i
v
i
de
d
i
n
t
o
t
h
r
e
e
s
t
a
g
e
s
,
p
r
e
p
r
o
c
e
s
s
i
n
g
,
f
e
a
t
u
r
e
e
xt
r
a
c
t
i
o
n
,
a
n
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
.
2.2.1. P
repr
oc
essing
The
first
ste
p
in
the
prep
r
oces
sing
sta
ge
is
norm
al
iz
ation
.
The
raw
sig
nal
s
we
re
norm
al
i
zed
int
o
ze
ro
m
eans
with
a
m
ini
m
u
m
a
mp
li
tud
e
of
-
1,
and
a
m
axi
m
um
a
m
pli
tud
e
of
1.
In
(1)
a
nd
(2)
wer
e
use
d
to
norm
al
iz
e the r
aw
EE
G si
gnal
.
=
(1)
=
{
|
m
a
x
(
)
|
|
ma
x
(
)
|
≥
|
min
(
)
|
|
min
(
)
|
|
max
(
)
|
<
|
min
(
)
|
(2)
The
sec
ond
ste
p
was
is
olati
ng
the
el
ect
r
oo
c
ulogram
(EO
G
)
sig
nal
from
t
he
EE
G
sig
nal.
The
E
O
G
sign
a
l
is
a
sign
al
gen
e
rated
by
ey
e
m
ov
e
m
ent
act
ivit
y
[23]
.
It
con
ta
ine
d
in
the
EEG
sign
al
duri
ng
t
he
data
colle
ct
ion
proc
edure. It
occur
s b
eca
us
e F
p1
was uti
li
zed as t
he
m
ai
n
chann
el
. Th
e
EO
G
si
gn
al
ge
ner
at
es
m
or
e
consi
der
a
ble
e
le
ct
rical
po
te
nt
ia
l
than
the
act
ual
EE
G
s
ign
al
.
The
refo
re,
it
ca
n
be
le
d
int
o
distor
ti
ng
inf
or
m
at
ion
.
The
m
os
t
com
m
on
m
et
ho
d
to
is
olate
the
EOG
sig
nal
is
ind
e
pe
ndent
com
po
ne
nt
a
na
ly
sis
(I
CA
).
Conver
sel
y,
ICA
require
d
m
ul
ti
-
channel
recorde
d
EE
G
sign
al
s
to
pr
oduce
an
e
xc
el
le
nt
resu
lt
.
As
this
researc
h
on
ly
us
e
d
a
s
in
gle
channel,
it
fou
nd
it
chall
eng
i
ng
to
isolat
e
the
EOG
si
gnal
usi
ng
ICA.
The
refor
e
,
this
stud
y
em
plo
ye
d
t
he
e
m
pirical
m
od
e
l
deco
m
po
sit
ion
(EM
D)
.
E
MD
dec
om
poses
signa
l
into
se
ver
al
intrinsic
m
od
e
fu
nc
ti
on
s
(I
M
Fs)
.
EM
D
w
orks
great
in
iso
la
ti
on
ta
sk
s
accor
ding
to
[
23]
.
Zahh
a
d
et
al
.
[23]
was
f
ound
that
the
EEG
si
gnal
con
ce
ntrated
in
the
first
a
nd
seco
nd
IMFs
.
H
ow
e
ver,
[23]
do
es
n’
t
s
pecify
the
m
axi
m
u
m
num
ber
of
dec
om
po
sed
IMFs
.
Ther
e
a
re
no
sign
ific
a
nt
ex
trem
a
a
t
the
11
th
IMF,
acc
or
ding
to
auth
or’s
obser
vation
ba
sed
on
c
ollec
te
d
sig
nal.
T
her
e
f
or
e
,
the
a
uthors
li
m
it
the
total
num
ber
of
decom
po
sed
IMFs
t
o
10
IM
Fs.
M
or
e
over,
it
found
t
hat
th
e
EE
G
sig
nal
c
on
ce
ntrate
d
only
in
the
first
I
MFs.
T
he
rem
ai
ni
ng
deco
m
po
se
d
si
gn
al
s
are
i
den
ti
fied
as
the
EO
G
sig
nal.
I
n
t
h
e
t
h
i
r
d
s
t
e
p
,
t
h
e
E
E
G
s
i
g
n
a
l
i
s
d
i
v
i
d
e
d
i
n
t
o
n
o
n
-
o
v
e
r
l
a
p
p
i
n
g
N
s
e
g
m
e
n
t
s
.
S
u
p
p
i
a
h
a
n
d
V
i
n
o
d
[
1
5
]
us
e
d
5
sec
onds
of
the
le
ng
t
h
of
se
gm
ents
to
div
ide
t
he
EE
G
sig
nal.
It
s
howe
d
a
high
c
orrelat
ion
betw
een
the
al
ph
a
ba
nd,
po
wer
s
pectral
de
ns
it
y
(P
SD)
fe
at
ur
e,
a
nd
acc
uracy
in
a
relaxing
sta
te
.
I
n
thi
s
stud
y,
5
sec
onds
of
the
segm
ent
w
ere
al
so
em
plo
ye
d.
T
her
e
fore
,
12
segm
ents
wer
e
e
xtracte
d
from
each
data.
The
al
pha
band
(8
-
12
Hz
)
was
extracte
d
f
ro
m
each
segm
ent
us
in
g
t
he
discr
et
e
wa
velet
tra
ns
f
or
m
(DWT
)
.
Da
ubechies
ty
pe
4
(db
4)
was
em
plo
ye
d
as
a
m
oth
e
r
of
wav
el
et
.
Fu
rth
erm
ore,
six
le
vels
of
deco
m
po
sit
io
n
wer
e
perfor
m
ed
to
ob
ta
in
the
al
ph
a b
a
nd. As a
r
e
su
lt
, 45x
12 d
at
a points
wer
e
c
ollec
te
d
in
this
ste
p.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
2
3
, N
o.
3
,
Se
ptem
ber
20
21
:
18
47
-
18
54
1850
2.2.2. F
ea
tu
re
s
ex
tra
c
tio
n
Fast
f
ourier
t
r
ansfo
rm
(F
FT)
base
d
on
the
Welch
m
et
hod
was
a
ppli
ed
t
o
est
im
at
e
PSD
[
25]
.
P
SD
est
i
m
ation
is e
xpresse
d
i
n
(
3)
.
̂
ℎ
(
)
=
1
∑
̂
−
1
=
0
(
)
(3)
̂
(
)
=
1
|
(
)
(
)
(
−
2
)
|
2
(4)
(
)
=
0
.
5
(
1
−
cos
(
2
)
)
,
0
≤
≤
(5)
̂
(
)
is
the
pe
rio
dogram
of
each
segm
ent,
w
he
re
=
1
,
…
,
−
1
are
data
s
egm
ents,
and
is
a
nu
m
ber
of se
gm
ents.
̂
(
)
est
i
m
ation
is g
i
ven
by
(
4).
is t
he
le
ng
th
of
the se
gm
ent.
is t
he
total
av
erag
e
of
(
)
.
(
)
is
the
wind
ow
functi
on
a
s
expresse
d
in
(5).
Ha
nnin
g
window
was
a
pp
li
ed
i
n
this
stud
y.
T
he
nu
m
ber
of
ove
rlap
ping
wi
ndows
us
ed
i
n
this
researc
h
is
50%.
A
t
otal
of
45x1
2x84
data
po
i
nts
we
re
e
xtracte
d
in the feat
ure e
xtracti
on sta
ge.
More
ov
e
r,
the
dim
en
sion
al
red
uctio
n
pro
ced
ur
e
wa
s
app
li
e
d.
Pr
i
ncipal
co
m
po
nen
t
analy
sis
(P
CA
)
and
li
near
disc
rim
inant
analy
sis
(L
DA)
a
re
know
n
as
a
de
qu
at
e
dim
ension
al
re
du
ct
io
n
m
e
tho
ds.
H
oweve
r,
PCA
wor
ks
we
ll
with
unla
be
le
d
data.
T
he
refor
e
,
L
DA
was
e
m
plo
ye
d
as
a
dim
ension
al
re
du
ct
io
n
m
et
hod.
As
a
sp
a
rse
so
l
ver,
sin
gula
r
valu
e
dec
om
po
sit
io
n
(SVD
)
a
nd
L
DA
m
axi
m
iz
e
the
dim
ension
a
l
reducti
on
m
eth
od
.
The dim
ensional
r
ed
uctio
n proced
ure re
duce
d
the
num
ber
of f
eat
ur
es
v
ect
or to
45x1
2x8.
2.2.3
. Cla
ssific
at
i
on
LDA
us
in
g
th
e
le
ast
-
sq
ua
re
so
lve
r
an
d
s
uppo
rt
vecto
r
m
achine
(SV
M)
us
in
g
li
ne
ar
ke
rn
el
we
re
app
li
ed
i
n
the
cl
assifi
cat
ion
s
ta
ge.
Cross
-
val
idati
on
w
as
em
plo
ye
d
to
sp
li
t
data
into
trai
ning
an
d
te
sti
ng
set
s.
The
nu
m
ber
of
=
10
was
em
plo
ye
d.
F
ur
t
her
m
ore,
the
cl
assifi
c
at
ion
sta
ge
wa
s
perform
ed
in
three
m
ai
n
conditi
ons.
T
he
first
c
onditi
on
is
cl
assifi
c
at
ion
by
usi
ng
the
com
bina
ti
on
of
al
l
segm
ents.
The
seco
nd
conditi
on
is
th
e
cl
assifi
cat
ion
of
each
se
gme
nt.
The
thi
rd
conditi
on
is
a
com
bin
at
ion
of
se
gm
ents.
In
th
e
com
bin
at
ion
of
segm
ents,
al
l
segm
ents
were
div
ide
d
int
o
three
sect
io
ns
with
an
e
qual
nu
m
ber
of
se
gm
ents,
the
first
sect
ion,
the
m
idd
le
sect
ion
,
a
nd
the
la
st
sect
i
on.
Each
sect
ion
c
on
sist
s
of
4
se
gm
ents.
The
com
bin
at
ion
of seg
m
ents f
r
om
the d
if
fer
e
nt
sect
ion
s
is als
o
em
plo
ye
d
i
n t
he
cl
assifi
cat
ion st
age
.
Table
1
s
hows
the
com
bin
at
ion
of
segm
ents
that
wer
e
us
ed
in
this
stud
y.
Group
G1,
G
2,
and
G3
a
r
e
the
gro
ups
f
r
om
the
segm
ents
in
the
first,
m
idd
le
,
an
d
la
st
sect
ion
,
res
pe
ct
ively
.
Group
G
4
c
om
bin
es
the
fir
st
segm
ents
from
each
sect
io
n.
Group
G
5,
G6,
an
d
G
7
fo
ll
ow
the
G
4
se
gm
ent
com
bin
at
ion
ru
le
,
al
though
t
hey
us
e
the
sec
on
d,
the
t
hir
d,
a
nd
t
he
f
ourth
segm
ents,
re
sp
ect
ively
.
Th
e
intenti
on
of
e
m
plo
ye
d
se
gm
ents
com
bin
at
ion
is
to
fin
d
a
s
pe
ci
fic
segm
ent
that
m
a
y
con
ta
in
pri
m
ary
in
form
ation
.
T
hus,
it
can
re
duce
the
com
pu
ta
ti
on
al
cost c
om
par
ed t
o
the
us
e
of al
l segm
ents.
Table
1
. Gr
oups o
f
se
gm
ent co
m
bin
at
ion
Grou
p
Na
m
e
Seg
m
en
t Co
m
b
in
a
tio
n
G1
1
,
2
,
3
,
4
G2
4
,
5
,
6
,
7
G3
9
,
1
0
,
1
1
,
1
2
G4
1
,
5
,
9
G5
2
,
6
,
1
0
G6
3
,
7
,
1
1
G7
4
,
8
,
1
2
3.
RESU
LT
S
AND DI
SCUS
S
ION
Figure
1
s
how
s
the
cl
assifi
cat
ion
res
ult
from
LDA
and
SV
M
us
i
ng
al
l
segm
ents.
78
%
and
70
%
accuracy
achie
ved
us
in
g
LD
A
an
d
SV
M,
r
especti
vely
.
Ba
sed
on
tho
se
accuracies
,
it
s
howe
d
unsat
isf
act
or
y
resu
lt
s
f
or
t
he
bio
m
et
ric
syst
e
m
.
Hen
ce,
the
analy
sis
based
on
eac
h
se
gm
e
nt
was
perfor
m
ed.
Fig
ur
e
2
sh
ows
the cla
ssific
at
ion res
ult t
hat e
m
plo
ye
d
each
segm
ent.
The
L
DA
sho
wed
bette
r
res
ults
com
par
ed
to
the
S
VM.
It
al
so
f
ound
th
at
the
aver
a
ge
accuracy
of
each
se
gm
ent
sh
owe
d
bette
r
accuracy
c
om
par
ed
to
the
acc
ur
acy
of
al
l
se
gm
ents.
The
L
DA
obta
ine
d
82.
58
%
accuracy,
al
th
ough
the
S
VM
ob
ta
ine
d
76.67
%
accuracy,
as
sh
ow
n
in
Fig
ure
3.
T
he
best
accuracy
wa
s
f
ound
at
segm
ent
num
ber
3
an
d
nu
m
ber
11
with
86%
acc
ur
acy
us
in
g
L
D
A,
as
show
n
i
n
Fi
gure
2.
N
onet
he
le
ss,
in
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c
En
g
&
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Id
e
ntif
ic
ation
of
o
ptim
um se
gment i
n
si
ng
le
chan
nel EE
G b
iometric
syste
m
(
Mu
ham
m
ad Afi
f Hendr
aw
an
)
1851
the
SV
M
cl
ass
ifie
r,
the
be
st
accuracy
al
so
le
ans
on
segm
ent
num
ber
3.
The
res
ult
ind
ic
at
ed
that
th
e
m
ai
n
inf
or
m
at
ion
m
i
gh
t
be
c
oncent
rated i
n
se
gm
e
nt num
ber
3.
More
ov
e
r,
t
he
perform
ance
of
eac
h
gro
up
of
seg
m
ent
com
bin
at
ion
s
was
inspe
ct
ed.
T
he
cl
assifi
cat
ion
r
esult
is
sh
own
in
Figure
4.
The
best
accu
ra
cy
was
ob
ta
ine
d
by
gro
up
G1
with
82%
accuracy
.
Ba
sed
on
Tabl
e
1,
G
1
c
on
sist
s
of
the
1
st
,
2
nd
,
3
rd
,
an
d
4
th
se
gm
ents.
The
re
su
lt
of
G
1
m
igh
t
be
i
nf
l
uen
ce
d
by
segm
ent
nu
m
ber
3.
Howe
ver,
the
fin
ding
of
se
gm
ent
num
ber
3
m
ay
i
ncr
ease
the
ac
cur
acy
of
G
1
was
not
fo
ll
owin
g
the
r
esult
of
G
6.
T
he
G
6
only
got
68
%
by
us
i
ng
LD
A
an
d
59
%
by
us
in
g
S
VM.
Lo
okin
g
furthe
r,
G6
c
onsist
s of
segm
ents 3
, 7, a
nd
11. As alre
ad
y sh
own
i
n Fi
gure
2,
se
gme
nts 3
a
nd
11
r
esulte
d
in
the
hi
gh
est
accuracy
us
in
g
LD
A.
T
hey
al
so
resu
lt
ed
in
t
he
fi
rst
an
d
se
cond
-
best
acc
uracy
us
i
ng
SVM
.
Be
sides,
se
gm
ent
nu
m
ber
s
7
a
nd
10
wer
e
the
le
ast
accurate.
It
al
so
found
that
segm
ent
nu
m
ber
10
was
the
s
eco
nd
le
ast
accuracy
with
79% acc
ur
acy
by u
si
ng L
DA.
Figure
1
.
Cl
assifi
cat
ion
resu
lt
from
all
seg
m
e
nts
Figure
2
.
Cl
assifi
cat
ion
resu
lt
for
eac
h
se
gm
e
nt
Figure
3
.
A
verage acc
ur
acy
from
all
seg
m
ents
Figure
4
.
Cl
assifi
cat
ion
resu
lt
from
each
gro
up
Fu
rt
her
m
or
e,
t
he
ne
w
gr
oups
of
se
gm
ent
com
bin
at
ion
s
wer
e
a
naly
zed
.
Table
2
s
ho
ws
the
new
com
bin
at
ion
of
segm
ents.
The
ne
w
gro
up
s
wer
e
create
d
based
on
the
best
segm
ent
accuracy
fro
m
each
sect
ion
.
Se
gm
e
nt
nu
m
ber
s
3,
5,
an
d
11
wer
e
the
best
segm
ents
from
each
sect
ion
,
res
pe
ct
ively
,
in
the
LDA,
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
2
3
, N
o.
3
,
Se
ptem
ber
20
21
:
18
47
-
18
54
1852
al
tho
ug
h
the
S
VM
m
ade
m
or
e
com
bin
at
ions
as
there
are
segm
ents
that
had
the
sam
e
accuracy
in
th
e
sa
m
e
sect
ion
.
T
he
ne
w
gro
ups
al
lo
wed
a
te
st
of
t
he
hypot
hesis
that
the
com
bi
nation
of
t
he
be
st
segm
ents
m
igh
t
increase
the
ac
cur
acy
.
T
he
hig
he
st
accuracy
was
achie
ved
by
G8
a
nd
G
11
us
i
ng
L
D
A,
as
sh
ow
n
in
Fi
gure
5.
Both
G
8
a
nd
G11
co
ns
ist
of
segm
ents
3
and
11.
H
owe
v
er
,
the
res
ult
of
G
8
a
nd
G
11
didn’t
e
xce
ed
the
accuracy
of
se
gm
ents 3
and 11.
T
her
e
f
or
e,
the f
eat
ur
es t
hat lea
n
on se
gm
e
nts 3
a
nd 11 ma
y be u
se i
n
th
e E
E
G
bio
m
et
ric syst
e
m
d
ue
to th
e
hig
he
st acc
uracy
.
The
st
ud
y
re
ve
al
s
the
possibil
it
y
of
si
ng
l
e
-
cha
nnel
EE
G
as
a
m
od
al
it
y
in
the
bio
m
et
ric
syst
e
m
,
especial
ly
in
the
resti
ng
ey
e
open
c
onditi
on.
The
pro
pose
d
m
et
ho
d
wa
s
m
or
e
s
uitable
in
real
conditi
ons
.
This
stud
y
al
so
pre
sents
a
se
gm
e
ntati
on
a
naly
sis
in
EE
G
pro
cessi
ng
as
a
bio
m
et
ric
m
od
al
it
y.
In
this
stud
y,
segm
e
nt
nu
m
ber
3
sho
wed
t
he
highest
acc
ur
acy
.
It
de
m
on
strat
es
that
th
e
propose
d
EE
G
bio
m
et
ric
req
ui
res
on
ly
15
seco
nds
to
disti
ngui
sh
on
e
pe
rs
on
from
oth
ers
with
rea
sona
ble
accura
cy
.
A
dd
it
io
nally
,
th
e
sm
a
ll
a
m
ou
nt
of
data u
se
d
i
n
the
pr
opos
e
d
m
et
ho
d m
a
y ca
us
e l
ower
co
m
pu
ta
ti
on
al
c
os
ts.
Table
2
. Ne
w gro
up
s
of se
gm
ent co
m
bin
at
ion
Grou
p
Na
m
e
Seg
m
en
t Co
m
b
in
a
tio
n
G8
3
,
5
,
1
1
G9
3
,
5
,
9
G1
0
3
,
8
,
9
G1
1
3
,
8
,
1
1
Figure
5
.
Cl
assifi
cat
ion
resu
lt
of the
ne
w
gro
up
s
of se
gm
ents
4.
CONCL
US
I
O
N
This
pa
per
st
udie
d
the
possibil
it
y
of
sing
l
e
-
cha
nnel
EE
G
as
a
bio
m
etr
ic
m
od
al
it
y.
The
pr
opos
e
d
m
et
ho
d
c
om
bi
ned
t
he
EO
G
rej
ect
io
n
an
d
segm
entat
ion
par
a
dig
m
to
pr
ovide
a
fai
r
pr
eci
sio
n
a
nd
reli
able
bio
m
et
ric
syste
m
.
The
EO
G
re
j
ect
io
n
m
et
ho
d
al
lo
we
d
the
data
colle
ct
io
n
pr
ocedur
e
t
o
be
pe
rfo
rm
ed
wh
il
e
the
sub
j
ect
w
a
s
resti
ng
with
ey
e
op
e
n
c
ondi
ti
on
.
It
was
cl
os
er
to
t
he
real
co
nd
it
io
n
wh
e
re
the
sub
j
ect
wasn’t
require
d
to
perform
any
oth
er
act
ion,
su
c
h
as
m
oto
r
i
m
aginar
y,
visu
al
i
m
aginar
y,
or
e
ven
preve
nt
bl
ink
i
ng
act
ivit
y.
The
segm
entat
ion
par
a
dig
m
sh
ows
that
the
m
ai
n
featur
es
m
igh
t
le
an
on
a
sp
eci
fic
s
egm
ent.
Ther
e
f
or
e,
w
e
are
a
ble
to
us
e
a
s
pecific
se
gm
ent
to
prov
i
de
a
hi
gh
-
qu
al
it
y
bio
m
et
ric
syst
e
m
.
Accurac
y
of
86%
was
ob
ta
i
ned
f
ro
m
15
s
econds
of
recorde
d
E
EG
.
It
took
l
onge
r
tha
n
a
c
onve
ntio
na
l
bio
m
et
ric
syst
e
m
,
su
c
h
as
a
fi
nge
rprint
syst
em
.
Howe
ver,
it
shows
that
EE
G
as
a
m
od
al
it
y
of
a
bi
om
et
ric
s
yst
e
m
is
pr
om
i
sing
t
o
pro
vid
e
a
hi
gh
-
secu
rity
syst
e
m
with
a
low
com
pu
ta
ti
on
al
cost.
I
n
the
fut
ur
e,
we
will
inv
e
sti
gate
the
sam
e
m
et
ho
d
on
a
la
rg
e
am
ou
nt
of
data.
The
overlap
ping
seg
m
ent
will
be
i
nv
e
sti
gated.
M
or
e
over,
EE
G
sign
al
consi
ste
ncy
is
a
m
ajo
r
iss
ue
i
n
the
stu
dy
of
bio
m
et
rics.
W
e
are
al
s
o
will
to
in
vestigat
e
t
hi
s
issue
reg
a
r
din
g
the
po
s
sibil
it
y of
t
he
EE
G si
gnal
that m
a
y chan
ge
as a
hum
an
ge
ts olde
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c
En
g
&
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Id
e
ntif
ic
ation
of
o
ptim
um se
gment i
n
si
ng
le
chan
nel EE
G b
iometric
syste
m
(
Mu
ham
m
ad Afi
f Hendr
aw
an
)
1853
REFERE
NCE
S
[1]
I.
Ja
y
ar
at
hne
,
M.
Cohen,
and
S.
Am
ara
kee
rthi,
“
Surve
y
of
E
EG
-
base
d
biometric
aut
hen
ti
c
at
ion
,
”
in
Proceedi
ngs
-
2017
IEE
E
8
th
I
nte
rnational
Co
nfe
renc
e
on
Awa
rene
ss
Sci
ence
a
nd
Technol
ogy
,
i
CAST
2017
,
vol.
2018
-
Janua
,
no
.
iCAS
T,
2017,
p
p.
324
–
329
,
doi
: 10.1109/
ICAw
ST.
2017.
8256471
.
[2]
A.
S.
Raj
u
and
V.
Uda
y
asha
nka
ra,
“
Biom
et
ri
c
p
erson
aut
hen
tication:
A
rev
i
ew,”
Proce
ed
ings
of
2
014
Inte
rnationa
l
Confe
renc
e
on
C
onte
mpor
ary
Co
mputing
and
In
f
orm
ati
cs,
IC3I
2
014
,
no
.
d
,
2014
,
pp.
575
–
580
.
[3]
K.
Dhara
vat
h
,
F.
A.
Ta
lukd
ar,
an
d
R.
H.
La
skar
,
“
Study
on
biom
et
ri
c
aut
h
ent
i
ca
t
i
on
s
y
stems
,
cha
l
le
nges
and
futur
e
tre
nds:
A
rev
ie
w
,
”
2013
I
EE
E
In
te
rnational
Conf
ere
nce
on
Computati
onal
Intelligence
and
Computing
R
ese
arch
,
IEE
E
ICCIC
20
13
,
2013
,
doi
:10.1109/ICCIC.
201
3.
6724278.
[4]
D.
Malt
oni
,
D.
Maio,
A.
K.
Jai
n,
and
S.
Prabh
aka
r,
"
Handboo
k
of
Fingerpri
nt
Rec
ognition,
"
2nd
ed.
London
:
Springer
Lo
ndon
,
2009
.
[5]
D.
A.
Re
y
nold
s
et
al.
,
“
The
2004
MIT
Li
ncol
n
La
bor
at
or
y
spea
ker
re
co
gnit
ion
s
y
st
em,”
ICASSP
,
IE
E
E
Inte
rnational
Co
nfe
renc
e
on
A
co
ustic
s,
Spe
ec
h
a
nd
Signal
Proc
e
ss
ing
-
Proce
ed
i
ngs
,
vol.
I
,
200
5
,
pp.
177
–
180
,
doi:
10
.
1109/IC
AS
S
P.2005.
1415079
.
[6]
J
.
D
a
u
g
m
a
n
,
“
H
o
w
I
r
i
s
R
e
c
o
g
n
i
t
i
o
n
W
o
r
k
s
,
”
T
h
e
E
s
s
e
n
t
i
a
l
G
u
i
d
e
t
o
I
m
a
g
e
P
r
o
c
e
s
s
i
n
g
,
v
o
l
.
1
4
,
n
o
.
1
,
p
p
.
7
1
5
–
7
3
9
,
2
0
0
9
.
[7]
T.
Matsum
oto,
H.
Matsum
oto,
K.
Yam
ada
,
and
S.
Hos
hino,
“
Im
pac
t
of
artifici
al
"
gum
m
y
"
fin
ger
s
on
finge
rpr
int
s
y
stems
,
”
Proceedi
ngs
of
SPIE
-
The
Inte
rnational
Soci
ety
for
Optic
al
Eng
ineeri
ng
,
vol.
4677,
n
o.
1,
pp.
275
–
289,
2002
,
doi
:
10
.
11
17/12.
462719.
[8]
K.
A.
Nixon
,
V
.
Aim
al
e, a
nd
R
.
K.
Rowe,
“
Spoof Det
e
ct
ion
Sche
m
es,
”
in
Handbook
of
Bi
om
et
rics
,
2007.
[9]
R.
B.
Nava
rro
,
L.
B
.
Vá
zque
z
,
and
E
.
L
.
Gui
llén,
“
EOG
-
base
d
whee
l
cha
ir
con
trol
,
”
in
Smar
t
Whe
elchairs
an
d
Brain
-
compute
r
Inte
rfac
es:
Mobi
le
Assistiv
e
Tech
nologi
es
,
2018.
[10]
C.
Rahmad,
R.
Ari
y
an
to,
and
D
.
R.
Yuniant
o
,
“
Brai
n
signal
cla
ss
ifi
ca
ti
on
using
Ge
net
ic
Algori
t
hm
for
right
-
le
ft
m
oti
on
pat
t
ern
,
”
Inte
rnational
J
ournal
of
Adv
a
nce
d
Computer
Sci
en
ce
and
Ap
pli
cations
,
vol
.
9,
no.
11
,
2018
,
doi:
10
.
14569/IJ
ACS
A.2018.
091134.
[11]
S.
Y.
Gordle
eva
et
a
l.
,
“
Real
-
Tim
e
EE
G
-
EMG
hum
an
-
m
ac
hine
int
er
face
-
base
d
cont
ro
l
s
y
st
em
for
a
lower
-
li
m
b
exoske
leton,
”
IE
EE
Ac
c
ess
,
vo
l.
8,
pp
.
84070
–
84
081,
2020
,
doi
:
1
0.
1109/ACCESS
.
2020.
2991812
.
[12]
I.
Odinak
a,
P.
H.
Lai,
A.
D
.
Kapla
n,
J.
A
.
O
’Sulli
van
,
E
.
J
.
Sireva
ag
,
and
J.
W
.
Rohrbaugh
,
“
ECG
biometric
rec
ogni
ti
on:
A
c
om
par
at
i
ve
an
aly
sis,
”
IEE
E
Tr
ansacti
ons
on
Information
Forensic
s
and
Sec
urit
y
,
vol.
7
,
no.
6,
pp.
1812
-
1824
,
2012
,
doi
:
10
.
11
09/T
IFS
.
2012.
2
215324.
[13]
S.
Marc
el
and
J.
del
R.
Mill
an,
“
Person
aut
hent
i
c
at
ion
using
bra
i
nwave
s
(EE
G)
a
nd
m
axi
m
um
a
posteri
ori
m
odel
ad
aptati
on
,
”
IE
E
E
Tr
ansacti
ons
on
Pat
te
rn
Analysis
and
Mac
hine
Inte
l
li
gen
ce
,
v
ol.
29,
no.
4
,
pp.
743
–
748,
2007
,
doi:
10
.
1109/T
P
AM
I.
2007.
1012.
[14]
M
.
D
e
l
P
o
z
o
-
B
a
n
o
s
,
J
.
B
.
A
l
o
n
s
o
,
J
.
R
.
T
i
c
a
y
-
R
i
v
a
s
,
a
n
d
C
.
M
.
T
r
a
v
i
e
s
o
,
“
E
l
e
c
t
r
o
e
n
c
e
p
h
a
l
o
g
r
a
m
s
u
b
j
e
c
t
i
d
e
n
t
i
f
i
c
a
t
i
o
n
:
A
r
e
v
i
e
w
,
”
E
x
p
e
r
t
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
4
1
,
n
o
.
1
5
,
p
p
.
6
5
3
7
–
6
5
5
4
,
2
0
1
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
s
w
a
.
2
0
1
4
.
0
5
.
0
1
3
.
[15]
R.
Suppiah
and
A.
P
.
Vinod,
“
Biom
et
ric
ide
n
ti
f
icati
on
using
singl
e
cha
nn
el
EE
G
during
rel
ax
ed
r
esti
ng
state,
”
IE
T
Bi
ometri
cs
,
vol
.
7,
no
.
4
,
pp
.
342
–
348,
2018
.
[16]
R
.
B
.
P
a
r
a
n
j
a
p
e
,
J
.
M
a
h
o
v
s
k
y
,
L
.
B
e
n
e
d
i
c
e
n
t
i
,
a
n
d
Z
.
K
o
l
e
s
,
“
T
h
e
e
l
e
c
t
r
o
e
n
c
e
p
h
a
l
o
g
r
a
m
a
s
a
b
i
o
m
e
t
r
i
c
,
”
C
A
N
A
D
I
A
N
C
o
n
f
e
r
e
n
c
e
o
n
E
l
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
2
,
p
p
.
1
3
6
3
–
1
3
6
6
,
2
0
0
1
,
d
o
i
:
1
0
.
1
1
0
9
/
C
C
E
C
E
.
2
0
0
1
.
9
3
3
6
4
9
.
[17]
R
.
P
a
l
a
n
i
a
p
p
a
n
a
n
d
D
.
P
.
M
a
n
d
i
c
,
“
B
i
o
m
e
t
r
i
c
s
f
r
o
m
b
r
a
i
n
e
l
e
c
t
r
i
c
a
l
a
c
t
i
v
i
t
y
:
A
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
P
a
t
t
e
r
n
A
n
a
l
y
s
i
s
a
n
d
M
a
c
h
i
n
e
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
2
9
,
n
o
.
4
,
p
p
.
7
3
8
–
7
4
2
,
2
0
0
7
,
d
o
i
:
1
0
.
1
1
0
9
/
T
P
A
M
I
.
2
0
0
7
.
1
0
1
3
.
[18]
G.
Schal
k,
D
.
J.
Mcfa
rla
nd
,
T
.
Hinte
rbe
rg
er,
N.
Birba
um
er,
and
J.
R.
W
olpa
w,
“
BCI2000 :
A
Gene
ral
-
Purpos
e
Brai
n
-
Com
pute
r
Inte
rfa
c
e
(BCI
)
Sy
st
em,”
IEEE
Tr
ansacti
ons
on
Bi
omedic
al
Engi
nee
ring
,
vol.
51,
no.
6
,
pp.
1034
–
1043
,
2004
,
doi
:
10
.
11
09/T
BME.2004
.
827072.
[19]
D.
La
Rocca
e
t
al.
,
“
Hum
an
bra
in
disti
n
ct
iv
e
ness
base
d
on
EE
G
spec
tr
al
c
oher
ence
connect
ivit
y
,
”
IEEE
Tr
ansacti
ons on Biomedical Engineering
,
vol
.
61
,
no.
9,
pp.
2406
–
2412,
2014
,
doi
: 10.1109/
TBME
.
2014.
2317881.
[20]
M.
Fraschi
ni
,
A.
Hill
ebr
and,
M
.
Dem
uru,
L.
Did
ac
i
,
and
G
.
L
.
Marc
ialis,
“
An
EE
G
-
base
d
bio
m
et
ric
s
y
s
te
m
u
sing
ei
genv
ec
tor
ce
nt
ral
ity
in
r
esti
ng
state
bra
in
net
wor
ks,”
IEE
E
Signa
l
Proce
ss
ing
Letters
,
vol.
22,
no.
6,
pp.
666
–
670
,
2015
,
doi
:
10
.
11
09/L
SP
.
2014.
23
67091.
[2
1]
K.
P.
Thomas
and
A.
P.
Vinod,
“
Biom
et
ric
ide
n
t
ifi
c
at
ion
of
p
ersons
using
sam
pl
e
ent
rop
y
f
ea
tu
r
es
of
EE
G
durin
g
rest
stat
e
,
”
201
6
IEE
E
Inte
rnat
ional
Confe
ren
c
e
on
Syste
ms
,
Man,
and
Cyber
net
ic
s,
SMC
2
016
-
Confe
renc
e
Proce
ed
ings
,
20
17
,
pp
.
3487
–
34
92,
doi
:
10
.
1109
/SM
C.
2016.
784
4773.
[22]
J.
H.
Kang,
Y.
C.
Jo,
and
S.
P.
Kim
,
“
El
e
c
troe
nc
epha
logr
a
phic
f
ea
tur
e
ev
al
ua
ti
on
for
improving
per
sona
l
aut
hen
ti
c
at
ion
p
erf
orm
anc
e
,
”
Ne
urocomputing
,
v
ol.
287
,
pp
.
93
–
1
01,
2018
,
doi
:
10
.
1016/j.ne
u
com.2018.
01.
074
.
[23]
M.
Abo
-
Za
hhad
,
S.
M.
Ah
m
ed,
and
S.
N.
Abb
as,
“
A
novel
bi
om
et
ric
appr
oa
c
h
for
hum
an
id
ent
ifica
ti
on
and
ver
ifica
t
ion
usin
g
e
y
e
bli
nk
ing
signal
,
”
I
EE
E
S
ignal
Pro
ce
ss
in
g
Letters
,
vol
.
22,
no.
7,
pp.
876
–
880,
2015
,
doi:
10
.
1109/L
S
P.2014.
2374338.
[24]
M.
Zey
n
al
i
and
H
.
Se
y
eda
r
abi,
“
EE
G
-
base
d
s
ingl
e
-
cha
nne
l
a
uthe
ntica
ti
on
s
ystems
with
optim
um
el
ec
trode
pla
c
ement
for
diffe
ren
t
m
en
ta
l
a
ct
iv
it
i
es,
”
Bi
omedic
a
l
J
ournal
,
vol.
4
2,
no.
4,
pp.
261
–
267,
2019
,
doi:
10
.
1016/j.bj
.
2019.
03
.
005.
[25]
P.
D.
W
el
ch,
“
T
he
Us
e
of
Fast
Four
ie
r
Tr
ansform
for
the
Esti
m
at
ion
of
Pow
er
Spect
ra
:
A
Meth
od
Based
on
Tim
e
Avera
ging
Ove
r
Short,
Modifi
ed
Periodogra
m
s,”
IEE
E
Tr
ansacti
o
ns
on
Audi
o
and
Elec
troacoust
ics
,
vol.
15,
no.
2,
pp.
70
–
73
,
1967
,
doi: 10.
1109
/TA
U.1967.
1161901.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vo
l.
2
3
, N
o.
3
,
Se
ptem
ber
20
21
:
18
47
-
18
54
1854
BIOGR
AP
HI
ES OF
A
UTH
ORS
Muhammad
Afif
H
endra
w
an
rec
e
ive
d
bac
h
elor
degr
e
e
from
Depa
rtment
of
Inform
at
ion
S
y
stems
,
Instit
ut
Te
knologi
Sepu
lurh
Nopem
ber
in
2015
and
m
aste
r
degr
e
e
from
Depa
rtment
of
El
e
ct
ri
cal
Eng
ine
er
ing,
Insti
tut
Te
knolog
i
Sepu
luh
Nopem
ber
in
2017.
His
cur
re
nl
y
working
as
assistant
prof
e
ss
or
at
Polit
ek
nik
Nege
ri
Mal
ang.
His
rese
ar
ch
int
er
est
is
in
the
area
of
art
if
ic
i
al i
nt
el
l
ig
enc
e
,
edge c
om
puti
ng,
signal pro
ce
ss
ing,
and
br
ain
-
computer
in
te
r
fac
e
.
Pr
amana
Yoga
Sap
utra
rec
ei
ved
b
ac
he
lor
degr
ee
from
Depa
rtment
o
f
Inform
at
ic
s
Engi
ne
eri
ng,
In
stit
ut
Te
kno
logi
Sepuluh
Nop
ember
in
2
010
and
m
aster
d
egr
ee
from
Depa
rtment
of
T
ec
hnolog
y
Mana
gement,
Instititu
t
Te
knologi
Sep
uluh
Nopem
ber
in
2014.
His
rese
arc
h
in
te
r
est
is
hu
m
an
-
computer
int
er
action,
art
ifi
c
ial
int
elli
genc
e
,
and
nat
u
ral
la
ngu
age
pr
oce
ss
ing.
Cah
y
a
Rah
mad
rec
e
ive
d
BS
de
gre
e
from
Brawi
jay
a
Univ
eri
sit
y
in
1998
and
MS
degr
ee
from
Depa
rtment
of
I
nform
at
ic
s
Eng
i
nee
ring
,
Insti
tut
Te
kno
logi
Sep
uluh
Nopem
ber
in
2005
.
He
rec
e
ive
d
a
doc
to
ral
deg
ree
from
Saga
Univer
sit
y
,
Japa
n,
in
2013
.
He
is
a
le
c
ture
r
i
n
Polit
ekn
ik
Nege
ri
Mal
ang.
His
rese
arc
h
intere
st
ar
e
image
proc
essing,
da
ta
m
ini
ng,
pat
t
ern
rec
ognition,
art
if
ic
i
al i
nt
el
l
ig
enc
e
,
and
br
ai
n
-
computer
in
te
rf
a
ce
.
Evaluation Warning : The document was created with Spire.PDF for Python.