TELKOM
NIKA
, Vol.11, No
.4, Dece
mbe
r
2013, pp. 74
9~7
5
8
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v11i4.1390
749
Re
cei
v
ed O
c
t
ober 8, 20
12;
Revi
se
d Oct
ober 2
9
, 201
2; Acce
pted
No
vem
ber 1
2
,
2012
Palmprint Verification Using Time Series Method
Agus
An
w
a
r
*
, Darma Putra, Agung
Cah
y
a
w
a
n
Dep
a
rtment of Information T
e
chno
log
y
, U
d
a
y
a
na U
n
ivers
i
ty
Bukit Jimbar
an
, Badung, Ba
li, Indon
esi
a
,
T
e
lp. 0361-
78
535
33
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: leke1
2
1
2
@g
mail.com
1
, ikgd
armap
u
tra@g
m
ail.com
2
,
agu
ng.ca
h
y
a
w
an@
gmai
l.com
3
Abs
t
rak
Peng
gu
naa
n
bio
m
etrik seb
aga
i siste
m
pen
ge
nal
an o
t
omatis berk
e
mb
an
g
pes
at
dal
a
m
peny
eles
ai
an
mas
a
l
ah ke
a
m
a
n
a
n
. Sala
h
satu yan
g
s
e
rin
g
di
gu
nak
an a
dal
ah
dat
a tela
pak ta
n
gan.
Pene
litia
n in
i me
ng
gun
aka
n
meto
de titik
momen d
ua tah
a
p
untuk se
gme
n
tasi reg
i
on
of interest (ROI) dan
me
ner
apka
n
meto
de
runt
u
n
w
a
ktu ya
n
g
d
i
ko
mb
inas
i
k
an
den
ga
n
meto
de
bl
ock
w
i
ndow
se
b
aga
i
repres
entasi
fit
u
r. Nor
m
ali
z
e
d
euc
lid
ea
n
dist
ance
di
gu
nak
a
n
u
n
tuk
meng
ukur k
e
sa
ma
a
n
d
u
a
vektor
fi
tur
telap
a
k ta
nga
n
.
Peng
uji
a
n
si
stem v
e
rifik
a
si
pa
da
pe
ne
liti
an
ini
me
ngg
u
naka
n
500
sa
mp
el
citra te
la
pak
tanga
n, den
ga
n 6 sa
mp
el se
bag
ai citra u
ji d
an 4 sa
mpel s
eba
gai citr
a ac
uan. Has
il p
e
n
guji
an
me
nu
nju
k
a
n
bahw
a siste
m
men
g
h
a
silk
a
n
unj
uk kerj
a tingg
i den
gan tin
g
kat keber
hasi
l
a
n
me
nca
pai 9
7
,33
%
(F
NMR=
1,67 %, F
M
R=
1,00 %, T
=
0,036).
Ka
ta
k
unc
i:
bi
ometrik, verifik
a
si siste
m
, runt
un w
a
ktu, meto
de bl
ock w
i
ndo
w
,
telapak ta
ng
an
A
b
st
r
a
ct
The use of biom
etrics as an
autom
atic rec
o
gnition system is growi
ng rapidly in solv
ing security
prob
le
ms; p
a
l
m
pr
int is
on
e o
f
bio
m
etric sys
tem w
h
ic
h o
fte
n use
d
. T
h
is
p
aper
use
d
tw
o steps in
cent
er o
f
mass
mo
ment
meth
od for regi
on of inter
e
st (ROI)
segme
n
tatio
n
an
d
apply the ti
me ser
i
es
me
thod
combi
ned
w
i
th
block w
i
ndow
meth
od
as fe
at
ure re
pres
enta
t
ion.
Nor
m
ali
z
e
d
Euc
lid
ea
n D
i
s
tance
is us
ed
to
me
asur
e the s
i
milar
i
ty de
gre
e
s of
tw
o feature vectors
of pal
m pr
int.
S
ystem testi
ng i
s
don
e usi
ng
50
0
sampl
e
s pal
ms
, w
i
th 4 sampl
e
s as the refer
ence i
m
ag
e a
nd the 6 sa
mp
les as test ima
ges. Experi
m
e
n
t
results sh
ow
th
is syste
m
ca
n
achi
eve
a h
i
gh
perfor
m
a
n
ce w
i
th success
rat
e
a
bout
97.3
3
%
(FNMR=1.6
7
%,
F
M
R=
1.00 %, T
=
0.036).
Ke
y
w
ords
: bio
m
etric, verific
a
tion syste
m
, time
series, bl
ock w
i
ndow
meth
o
d
, pal
mpr
i
nt
1. Introduc
tion
The nee
d for personal
re
cognition
syst
em is
autom
atically a reli
able an
d tru
s
two
r
thy
increa
sing m
a
inly
for se
curity
system
s. Recognitio
n
system
aim
s
to solve
a p
e
rson’
s ide
n
tity.
There are two type
s of
recogni
tio
n
system
s, the verificatio
n
and id
enti
f
ication
syst
ems.
Verificatio
n
system aim
s
to accept
or reje
ct
the
claime
d ide
n
t
ity of a person,
while
the
identificatio
n system aim
s
to solve the id
entity.
Palmprint ha
s seve
ral a
d
vantage
s com
pare
d
to
othe
r available fe
ature
s
: low-re
solutio
n
image
s ca
n b
e
use
d
, low cost ca
pture d
e
vice
s ca
n b
e
used, it is very difficult or impossible to
Che
a
t with p
a
lmpri
n
t be
ca
use th
eir
ch
a
r
acte
ri
stics a
r
e stabl
e an
d
uniqu
e [1]. Many app
roa
c
h
e
s
have bee
n propo
sed fo
r extract the
pal
m
p
rint features
[1-11]. The
r
e
are ma
ny uni
que featu
r
e
s
in
a palmp
rint image that ca
n be u
s
ed fo
r person
a
l re
cognition, such us: ge
omet
ry, princi
pal li
nes,
wrin
kle
s
, delt
a
points a
nd
minutiae poi
n
t
s [2].
This pa
per
apply time
seri
es for p
a
lmpri
n
t feat
ure
extration
and
representation.
Basically, a ti
me
seri
es i
s
a colle
ction
o
f
obser
vatio
n
s
m
ade
sequ
entially in tim
e
. Neve
rthel
e
ss,
as l
ong
a
s
th
e data
of inte
rest
can
be
repre
s
e
n
ted seque
ntially,
time seri
es
technolo
g
ies ca
n
b
e
applie
d [12]. We u
s
ed two
steps in
cent
er of mass
m
o
ment metho
d
for regi
on o
f
interest (RO
I
)
segm
entation
and ap
ply the time se
rie
s
method
com
b
ined
with blo
c
k wi
ndo
w met
hod a
s
featu
r
e
rep
r
e
s
entatio
n an
d n
o
rm
a
lized
eu
clide
an di
stan
ce
i
s
a
dopte
d
a
s
the
mat
c
hi
ng
score
of t
w
o
templates
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 749
– 758
750
2. Res
earc
h
Method
The ve
rificat
i
on p
r
o
c
e
s
s ha
s fou
r
main
stage
s there
a
r
e
image
a
c
quisitio
n
,
prep
ro
ce
ssin
g, feature
ex
traction
an
d
matchin
g
.
Ov
erview of the
verificatio
n
pro
c
e
s
s in th
is
study
can
be
see
n
in
Figu
re 1. In g
ene
ral, a pal
mp
ri
n
t
verification
system co
nsi
s
ts of
4
m
odu
les
is
as
follows
:
1.
Stages of im
age a
c
qui
siti
on, a step to
take
biom
etric d
a
ta su
ch
as the imag
e of the
palmp
rint fro
m
the hand.
2.
Preprocessin
g
is the
stag
e
s
to t
r
an
sform the
origi
nal
image
into
g
r
eyscale
ima
ge, bin
a
ry
and ROI seg
m
entation (re
g
ion of intere
st) t
hat use for feature extraction p
r
o
c
e
s
s.
3.
Stages fe
atures
extractio
n
is th
e sta
g
e
s
to
determin
e
the
ch
ara
c
t
e
risti
c
featu
r
e
s
of th
e
segm
entation
that come fro
m
the value of tested ROI palmp
rints.
Matchin
g
sta
ges, nam
ely stage
s to det
ermin
e
t
he de
gree of
simila
rity between t
e
sted bi
ometric
traits
with biometric
features
that referenc
e
to the data
base, to deci
de wh
ether th
e use
r
is
accepte
d
or rejecte
d
ba
se
d on matchi
n
g
of test sco
re results.
2.1 Hand
Image
Acquisi
tion
Palmprint im
age
s a
r
e
cap
t
ured
usi
ng d
i
gital ca
mera
on
sma
r
t ph
one Sam
s
u
n
g
Galaxy
GT-S5
830
with re
solution
o
f
640x480 pix
e
ls. Each pe
rson
wa
s re
qu
ested to p
u
t his/her l
e
ft han
d
palm d
o
wn o
n
with
a
bla
c
k b
a
ckg
r
ou
nd
(ten
sample
for ea
ch
pe
rson). T
h
e
r
e a
r
e some
peg
s on
the board to control the ha
n
d
oriente
d
, transl
a
tion, an
d stret
c
hing a
s
sh
own in Figure 2.
Gr
a
ysc
a
l
in
g
B
i
nar
i
s
at
i
o
n
Ot
s
u
M
e
t
h
o
d
RO
I
t
w
o
s
t
e
p
i
n
mo
me
n
t
c
e
ntr
a
l
m
e
t
hod
In
t
e
n
s
i
t
y
N
o
r
m
a
liz
a
t
io
n
S
p
i
r
al
A
r
c
h
i
m
ede
s
F
e
a
t
u
r
N
o
r
m
a
liz
a
t
io
n
B
l
oc
k
W
i
nd
ow
M
e
t
hod
I
m
a
g
e ac
qui
s
i
t
i
o
n
P
r
e
p
r
o
ce
ss
i
n
g
F
eat
ur
e
E
x
tr
ac
t
i
o
n
Da
t
a
b
a
s
e
Ve
r
i
f
i
c
a
t
i
o
n
Ve
r
i
f
i
c
a
t
i
o
n
N
o
r
m
a
liz
e
d
E
u
c
l
i
d
ea
n
D
i
st
an
ce Met
h
o
d
Re
s
u
l
t
au
tho
r
i
z
ed
/
unau
tho
r
i
z
ed
2.2
Segmenta
tio
n
of Palmprint ROI
This
pap
er
used ne
w te
ch
n
i
que to
extract the ROI i
s
called two
ste
p
s in
mom
ent
cent
ral
method. The
main prin
cip
l
e of this me
thod is to
fix the locatio
n
of the moment cente
r
which
improve
s
the
palm ce
nter point locatio
n
[1]. T
he first step i
s
to determin
e
the ce
nter p
o
i
nt
moment
s
(
,
)
in the binary image by usi
ng equatio
n (1). Otsu me
thod has be
en use
d
to
binary ima
ge
[12]. Where
(
,
)
represents
cente
r
of are
a
is define
d
a
s
Equation 1.
1
,
1
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
9
30
Palm
print Verification Usin
g Tim
e
Series Method
(Ag
u
s Anwar)
751
Whe
r
e
N rep
r
esent nu
mb
er of pixel
s
obje
c
t. T
he centroid
of the se
gmente
d
binary ima
g
e
is
comp
uted
an
d ba
sed
on
th
is
centroid th
e ROI of g
r
ay
scale p
a
lmp
r
i
n
t image
can
be
cro
p
pe
d with
size 128x1
2
8
and 64x64 pi
xels.
2.3 Palmprint
Normalization
The n
o
rm
alization p
r
o
c
e
s
s i
s
ne
ede
d
to re
du
ce t
he p
o
ssible
imperfe
ction
s
in the
palmp
rint im
age d
ue to
n
on-u
n
iform
ill
umination
[1
4]. The n
o
rm
alizatio
n met
h
od e
m
ploye
d
in
this re
sea
r
ch
using eq
uati
on (2
). Whe
r
e
I
and
I’
represent origi
n
al grayscal
e palmp
rint ima
g
e
and the no
rm
alize
d
image
respe
c
tively,
ϕ
and
ρ
rep
r
ese
n
ts me
an
and varia
n
ce of the origi
n
al
image rep
e
cti
v
ely,
while
ϕ
d
and
ρ
d
a
r
e t
he d
e
si
red
value
s
for me
an an
d vari
a
n
ce
re
sp
ectiv
e
ly.
This re
sea
r
ch
use
ϕ
d
= 1
75 an
d
ρ
d
=
175 fo
r all
experim
ents,
and the
re
sults a
r
e
sho
w
n in
Figure (3
).
′
,
,
(2)
,
(3)
a
b
c
Figure 3. (a, b, c) Norm
alization results with
ϕ
d
= 175 a
nd
ρ
d
= 175
Figure 4. Use
Spiral as the
Feature Sequ
ence
Extrac
tion Track
2.4
Palmprint Fe
ature Ex
tra
c
tion
For time se
ri
es re
pre
s
e
n
tation purpo
se, ROI squa
res is de
com
p
osed into seque
ntial
data. In this rese
arch
we a
dopt a spi
r
al
as t
he tra
c
k for de
com
p
o
s
i
t
ion. The pol
a
r eq
uation o
f
a
spiral equ
atio
n 4 [12].
(4)
Whe
r
e a i
s
set to 0.7. Features
are extract
ed al
ong t
he spi
r
al (Fig
ure 4
)
. Many kind
s of
local textural
feature
s
ca
n be use
d
here
,
such a
s
intensity, varian
ce, and cro
s
s co
rrelation.
In
this pap
er, we use g
r
ay le
vel intensity.
′
(5)
For
ea
ch p
o
i
n
t on the
spi
r
al, the l
o
cal
textural featu
r
e extracte
d to form th
e d
a
ta seque
n
ce
S
whi
c
h
usually ha
s a
quite
h
i
gh dim
e
n
s
io
n a
s
sho
w
on
i
n
Figu
re
5
(a). This l
e
ad
s t
o
computatio
nal
heavy enou
g
h
, we ad
opt
block wi
ndo
w metho
d
to solve this
probl
em (Eq
u
ation 5
)
. Data
seq
uen
ce
repres
entation into
usin
g
block
wind
ow metho
d
. Figure 5 (b) sh
ow the
norm
a
lized fe
ature
s
usi
ng
block wi
ndo
w method.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 749
– 758
752
(a)
(b)
Figure 5. (a)
Data Sequ
en
ce of Featu
r
e
Extrac
tion, (b) Data Se
qu
ence of Featu
r
e No
rmli
zati
on
with Block Wi
ndo
w Method
2.5 Matchin
g
Matchin
g
bet
wee
n
the te
st
image
with t
he refere
nce
image
usin
g
Normali
z
ed E
u
clid
ean
Dis
t
ance [14]:
̅
,
̅
(6)
‖
‖
‖
‖
(7)
with
u
and
v
is a vecto
r
of traits characteri
stic
of
test imag
es a
nd the refe
re
nce im
age. T
h
e
smaller
̅
,
sco
re, the more
simila
r two fe
ature ve
ctors are mat
c
he
d
,
otherwi
se t
he high
er
the sco
r
e, th
e mo
re
different the
two
chara
c
te
rist
i
c
vectors. To
d
e
termin
e
whe
t
her th
e te
st i
s
a
valid image
(genui
ne)
or i
n
valid (Imp
oster) u
s
e
d
a
th
reshold val
u
e
.
When
the di
stan
ce
(sco
re
) i
s
smalle
r than
the thresh
ol
d value then
the user
i
s
said to be a
u
thori
z
ed, if large
r
then said
unauth
o
ri
zed use
r
s.
3. Resul
t
s
and
Analy
s
is
Verificatio
n
system whi
c
h
made a
bout
500 ima
g
e
s
were teste
d
usin
g the pal
m of the
hand bel
ong
s to 50 people
,
each pe
rso
n
rep
r
e
s
ente
d
10 sam
p
le image
s palm
s
. Six of
the 10
sampl
e
s
we
re use
d
a
s
test image
s, while 4 imag
e use
d
for the
referen
c
e im
age. Te
sting
the
su
ccess rate
of
syste
m
verification
i
s
do
ne
by
han
d
f
o
r
valu
e
F
M
R (Fal
se Mat
c
h Rate
),
F
N
MR
(Fal
se Non Match Rate),
and EER (Equal Error Ra
te). FMR is the matchin
g
error when
the
system i
s
co
nsid
ere
d
a d
i
fferent imag
e with the re
feren
c
e ima
g
e
is a sampl
e
image of t
he
partici
pant
s b
e
long to the
same
han
d. FNM
R
is
the
matchin
g
error when the
system a
s
su
mes
the same
refe
ren
c
e
imag
e
with the
sam
p
le ima
ge i
s
the ima
g
e
of t
he h
and
bel
o
ngs to a
different
partici
pant. EER is the erro
r rate when F
N
MR
= FM
R.
The same m
a
tchin
g
sco
r
e
(gen
uine
score
)
obtain
e
d
from matchi
ng the sa
mpl
e
image
with all the f
eature
s
of th
e refe
ren
c
e i
m
age t
hat h
a
ve the sam
e
identity. Di
fferent match
i
ng
score
s
(im
p
o
s
ter
score
)
o
b
tained by m
a
tchin
g
the sample ima
g
e
with all the referen
c
e im
a
g
e
that has a different ide
n
tity.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Palm
print Verification Usin
g Tim
e
Series Method
(Ag
u
s Anwar)
753
3.1
Number o
f
Refer
e
nce T
e
s
t
This te
st is u
s
ed to an
alyze accuracy o
f
the system again
s
t the n
u
mbe
r
of ref
e
ren
c
e
s
that use
in th
is sy
stem. Databa
se
size
that us
e
d
in t
h
is te
st is
50
parti
cipant
s
with ROI 64x
64
pixels an
d 12
8x128 pixel
s
. Table 1 an
d 2
sho
w
s the
re
sult of this test.
Table 1. Nu
m
ber of refe
ren
c
e test with
ROI 64x64 pixels
Number of
Reference
Treshold
FNMR (%)
F
M
R
(%
)
Accur
a
cy
(%)
1 0.037
11.00
2.33
86.67
2 0.037
8.00
2.88
89.12
3 0.033
7.33
1.37
91.29
4 0.033
6.33
1.59
92.07
Table 2. Nu
m
ber of refe
ren
c
e test with
ROI 128x128 p
i
xels
Number of
Reference
Treshold
FNMR (%)
F
M
R
(%
)
Accur
a
cy
(%)
1
0.040
4.00
2.28
93.72
2
0.036
3.67
0.82
95.51
3
0.036
2.33
0.92
96.75
4
0.036
1.67
1.00
97.33
Figure 6. System accuracy
again
s
t num
ber of refe
ren
c
e
Table 1
and
2 ca
n be
pre
s
ente
d
with
a
cha
r
t a
s
sho
w
n in
Figu
r 6.
Line
with dot
mark in
Figure
6 sho
w
s syste
m
a
c
curacy with ROI
64x64 pi
xels a
nd lin
e
with tria
ngle
system
accu
racy
with ROI 128
x128 pixels.
Figure 6 sho
w
s th
at
the system accu
ra
cy increa
se
s
along
with th
e
numbe
r of ref
e
ren
c
e
s
in da
tabase.
3.2
Databa
se Size Tes
t
ing
This te
st i
s
usin
g vari
ou
s datab
ase
si
ze, they a
r
e
15, 25
an
d
50 p
a
rti
c
ipa
n
ts. The
system
wa
s
tested
with 4
referen
c
e
s
. Re
sult of thi
s
test i
s
sho
w
n in fig
u
re
7 - 12
with t
he
followin
g
note
:
FNMR is
sh
own by a blu
e
line and FM
R is sho
w
n b
y
a red line.
3.2.1
Sy
stem accu
rac
y
against 15 particip
a
n
ts
There are 3 d
a
taba
se with
15 parti
cipa
nts.
Figure 7 an
d Figure 8 sh
ows simul
a
tio
n
system
with 1
5
partici
pant
s.
70
75
80
85
90
95
100
1234
A
ccurac
y
(%
)
Number of Reference(s)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 749
– 758
754
Figure 7. Gra
ph FNM
R
/FM
R
agai
nst 15
partici
pant
s ROI 64 x 64 (a) databa
se 1, (b)
databa
se 2, (c) data
b
a
s
e 3
Figure 8. Gra
ph FNM
R
/FM
R
agai
nst 15
partici
pant
s ROI 128 x 128 (a) d
a
taba
se
1,
(b) d
a
taba
se
2, (c) d
a
taba
se 3
1.
Testing
the
system on
dat
aba
se 1. T
h
is test
obtai
ned
that syste
m
accuracy
with
ROI 6
4
x64
pixels i
s
9
6
.5
1% at thre
sh
old value
0.0
35 wi
th
F
N
M
R
i
s
1.11%
a
nd FM
R i
s
1.
59% (Fi
gure
7(a
)). while the cu
rrent te
st is u
s
ing
ROI 128x128
pixels, the sy
stem a
c
cura
cy is 100%
at
tresh
o
ld valu
e 0.003
with
FNM
R
0% an
d FMR 0% (F
igure 8
(
a
)).
2.
Testing
the
system on
dat
aba
se 2. T
h
is test
obtai
ned
that syste
m
accuracy
with
ROI 6
4
x64
pixels i
s
9
1
.8
2% at thre
sh
old value
0.0
31 wi
th
F
N
M
R
i
s
7.78%
a
nd FM
R i
s
0.
32% (Fi
gure
7(b
)).
while t
he current te
st is u
s
in
g RO
I 128x12
8
pixels, the
system a
c
curacy is 95.56%
at
tresh
o
ld valu
e 0.036
with
FNM
R
4.44%
and FMR 0%
(Figure 8(b)).
3.
Testing
the
system on
dat
aba
se 3. T
h
is test
obtai
ned
that syste
m
accuracy
with
ROI 6
4
x64
pixels i
s
9
1
.2
6% at thre
sh
old value
0.0
33 wi
th
F
N
M
R
i
s
7.78%
a
nd FM
R i
s
0.
87% (Fi
gure
7(c)).
whil
e the current te
st is
usi
ng
ROI
128x12
8
pixels, the
sy
stem a
c
cu
racy is 99.2
1
%
at
tresh
o
ld valu
e 0.038
with
FNM
R
0% an
d FMR 0.79%
(Figure 8(c)).
3.2.2
Sy
stem accu
rac
y
against 25 particip
a
n
ts
There a
r
e
2
databa
se
wit
h
15
pa
rticip
ants.
Fig
u
re
9 an
d Fig
u
re
10
sh
ows
si
mulation
system
with 2
5
partici
pant
s.
1.
Testing
the
system on
dat
aba
se 1. T
h
is test
obtai
ned
that syste
m
accuracy
with
ROI 6
4
x64
pixels is
82.5
8
% at thresh
old value 0.0
35 with F
N
M
R
is 5.3
3
% and FM
R is
2.08 % (Figu
r
e
9(a
)).
while t
he current te
st is u
s
in
g RO
I 128x12
8
pixels, the
system a
c
curacy is 97.17%
at
tresh
o
ld valu
e 0.033
with
FNM
R
2.67%
and FMR 0.1
7
% (Figu
r
e 1
0
(a
)).
2.
Testing
the
system on
dat
aba
se 2. T
h
is test
obtai
ned
that syste
m
accuracy
with
ROI 6
4
x64
pixels i
s
9
2
.8
7% at thre
sh
old value
0.0
31 wi
th
F
N
M
R
i
s
6.71%
a
nd FM
R i
s
0.
42% (Fi
gure
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Palm
print Verification Usin
g Tim
e
Series Method
(Ag
u
s Anwar)
755
9(b
)).
while
the
curre
n
t te
st is u
s
ing
ROI 128x
1
28 p
i
xels,
the syst
em
a
c
curacy is
9
8
.17%
at
tresh
o
ld valu
e 0.036
with
FNM
R
0.67%
and FMR 1.1
7
% (Figu
r
e 1
0
(b
)).
Figure 9. Gra
ph FNM
R
/FM
R
with 25 p
a
rticipant
s ROI
64 x 64 (a) d
a
taba
se 1, (b
) databa
se 2
Figure 10. Graph FNMR/F
MR agai
nst 2
5
partici
pant
s ROI 128 x 128 (a
) datab
a
s
e 1, (b
)
databa
se 2
3.2.3
Sy
stem accu
rac
y
against 50 particip
a
n
ts
There is 1 dat
aba
se with 5
0
partici
pant
s. Figure 11 a
nd Figu
re 12
sho
w
s simul
a
tion
system
with 5
0
partici
pant
s.
Testing the
system on dat
aba
se with 5
0
parti
ci
pant
s. This test obtained that
system
accuracy
with ROI 6
4
x64
pixels is
92.
07% at
thre
shold value
0.033
with FNMR is
6.33
% and
FMR is 1.5
9
% (Figure 1
1
). while the
current
test
is usi
ng ROI 128x128 pix
e
ls, the syst
em
accuracy is 9
7
.33%
at tre
s
hold value 0.
036 with F
N
MR 1.67 % a
nd FMR 1.0
0
% (Figure 12).
3.2.4
Comparis
on of the a
ccur
a
c
y
s
y
stem
Comp
ari
s
o
n
of the accura
cy system wit
h
va
riation da
tabase si
ze can be se
en in
Table 3
for ROI 64x6
4
pixels an
d
Table 4 for ROI 128x12
8 pixels, whil
e its gra
phics are sh
own
in
Figure 13.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 749
– 758
756
Table 3. System perfo
rma
n
ce
with RO
I
64x64 pixel
s
on variou
s da
tabase zi
se
Database
size
(in user)
DB
Treshold
FNMR
FMR
Accur
a
cy
Ti
me
(second)
15
1
0.035
1.11
2.3
96.59
6.36
15
2
0.031
7.87
0.32 91.82
6.40
15
3
0.033
7.87
0.87 91.26
6.45
25
1
0.035
5.33
2.08 92.58
16.66
25
2
0.029
6.71
0.36 92.93
17.41
50
1
0.033
6.33
1.59 92.07
61.51
Table 3
sho
w
s the
re
sult
an avera
ge
accuracy
of
92.87% with
a maximum
matchin
g
time on
testing
50
p
a
rticip
ants i
s
61.51
se
co
nds. V
e
rifi
c
a
t
i
on
sy
st
em
can
be
said
t
o
b
e
v
e
ry
f
a
st
matchin
g
of time less than
1 se
con
d
on
each parti
cip
ant
.
Table 4. System perfo
rma
n
ce
with ROI
128x12
8 pixe
ls on vaiou
s
d
a
taba
se zi
se
Database
size
(in user)
DB
Treshold
FNMR
FMR
Accur
a
cy
Ti
me
(second)
15
1
0.03
0
0
100
6.38
15
2
0.036
4.44
0
95.56
6.40
15
3
0.038
0
0.79
99.21
6.45
25
1
0.033
2.67
0.17
97.17
17.11
25
2
0.036
0.67
1.17
98.17
16.82
50
1
0.036
1.67
1
97.33
65.40
Table 4
sho
w
s the
re
sult
an avera
ge
accuracy
of
97.90% with
a maximum
matchin
g
time on
testing
50
p
a
rticip
ants i
s
65.40
se
co
nds. V
e
rifi
c
a
t
i
on
sy
st
em
can
be
said
t
o
b
e
v
e
ry
f
a
st
matchin
g
of time less than
1 se
con
d
on
each parti
cip
ant.
Figure 13
sh
own
both
ROI 64x64
an
d 128x1
28,
t
he pe
rforma
nce
(a
ccuracy) of the
system is
rel
a
tively stable
although si
ze of dat
aba
se increa
sed,
its m
ean that
system is n
o
t
affected by si
ze or d
a
taba
se or num
ber
of particip
ant
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Palm
print Verification Usin
g Tim
e
Series Method
(Ag
u
s Anwar)
757
Figure 13. Co
mpari
s
o
n
of system accu
ra
cy
4.
Con
c
lusi
on
Based
on
the
experi
m
ental
re
sults, th
e p
a
lmpr
i
n
ts bio
m
etrics
ve
rification system has
the
best p
e
rfo
r
m
ance on
the
ROI of 1
28x1
28 pixel
s
, proved value
F
N
MR/F
MR
sy
stem i
s
small
or
has
a high
degree of
accuracy,
with an a
ccu
ra
cy of rat
e
app
roxima
tely 97.33% or
FNM
R
=1.67
%
and FM
R=1%. Hig
h
p
e
rform
a
n
c
e
obtaine
d in this
study is
also
due to
the
prep
aration p
r
ocess of the
sepa
ratio
n
o
f
the pr
o
c
e
ssi
ng of palmp
ri
nt with two
steps in m
o
me
nt
central m
e
th
od a
nd im
ag
e inten
s
ity n
o
rmali
z
atio
n palms. The
verification
system
is
hig
h
ly
feasibl
e
to d
e
v
elop in th
e d
i
rectio
n of a
n
onli
ne
syste
m
and
applie
d t
o
sp
ecifi
c
a
p
p
licatio
n field
s
,
su
ch a
s
att
enda
nce sy
stems ap
plica
t
ions, m
edi
cal
re
co
rd
s appli
c
ation, se
curity
sy
st
e
m
appli
c
ation
s
, foren
s
i
c
appli
c
ations, an
d other ap
plicatio
ns.
Referen
ces
[1]
I Ketut Gede
Darma P
u
tra,
Erdia
w
a
n
. H
i
g
h
Perf
orma
nce
Palmpr
int Ide
n
tificatio
n
S
y
st
em Base
d O
n
T
w
o D
i
mens
io
nal Gab
o
r.
T
E
LKOMNIKA Journal of Electric
al Eng
i
ne
eri
ng.
2010; 8(3): 3
0
9
-31
8
.
[2]
Shu W
,
Z
hang
D. Automated
Person
al Ide
n
t
ification
b
y
Pa
lmprint.
Optica
l
Engi
neer
in
g.
199
8; 37(
8):
235
9 -23
63.
[3]
W
u
X Q, Z
hang D, W
ang K Q. F
i
sherpalms
Based Pa
lmpri
n
t Recog
n
itio
n.
Pattern Recog
n
itio
n Letters
.
200
3; 24: 282
9-28
38
[4]
Kong
W
K, Z
h
ang
D, L
i
W
.
P
a
lmpri
n
t
F
eatur
e E
x
tractio
n
us
ing
2-D G
abor
F
ilters.
Pattern
Reco
gn
ition.
200
3; 36: 233
9-23
47.
[5]
W
u
X Q, W
ang K Q, Z
h
a
n
g
D. W
a
ve
let
Based
Palm
pri
n
t Rec
ogn
itio
n
. Procee
din
g
s
of the F
i
rst
Internatio
na
l C
onfere
n
ce o
n
Machi
ne Le
arn
i
ng a
nd Cy
bern
e
tics.
2002; 3:
125
3- 12
57.
[6]
Li W
,
Z
hang D
,
Xu Z
.
Palmpr
in
t Identific
atio
n b
y
F
ouri
e
r T
r
ansform.
Internatio
nal J
ourn
a
l of Patte
r
n
Reco
gniti
on Ar
tifici
al Intelligence
. 2002; 1
6
(4
): 417-43
2.
[7]
Z
hang D, Shu
W
.
T
w
o Nov
e
l
Char
acteritics i
n
Palm
pri
n
t Ve
rificatio
n
: Datu
m Po
int Invaria
n
ce an
d Lin
e
F
eature Matchi
ng.
Pattern Re
cogn
ition.
1
998
; 32: 691-7
02.
[8]
Liu G M, Z
han
g D, W
a
n
g
K Q. Palmprint Rec
o
g
n
iti
on Usi
ng E
i
g
enp
alms F
eat
ures.
Pattern
Reco
gniti
on L
e
tters
. 200. 24: 146
3 – 14
67.
[9]
Duta N, Jain A
K, Mardia
K.V. Matching of Palmprint.
Pattern Recognation Letters.
2002; 2
3
: 477-48
5.
[10]
W
u
X, W
and K, Z
hang D.
F
u
zz
y
Dir
ectio
nal
El
ement E
nerg
y
F
eatur
e
(F
DEEF
) Based Palm
print
Identificati
on.
p
r
ocee
din
g
s 1
6
th
Internation
a
l
Confer
ences
o
n
pattern r
e
cog
n
itio
n,
Queb
ec.
200
2; 1: 95-
98.
[11]
I Ketut Darma
Putra, W
i
ra Bhua
na, Erdi
a
w
a
n
. Pem
bent
ukan K
ode T
e
lap
a
k T
angan
(Palm Co
de)
Berbas
is Meto
de Gabor 2
D
.
MAKARA Jour
nal of Technology Serie.
201
1; 15(2): 16
1-1
67.
[12]
Jian S
hen
g C
hen, Yi
u Sh
an
g Moo
n
, Hoi
W
o
Yeun
g. Palmpri
n
t Authe
n
ticatio
n
Usi
n
g
T
i
me Series
.
Dep
a
rte
m
ent o
f
Science an
d
Engi
neer
in
g T
he Chi
nese U
n
iv
ersity of Hongk
ong
. 20
05: 37
6
-
385.
[13]
Darma Putra.
Bineris
asi C
i
tra
T
angan d
e
n
g
a
n
Metode Otsu
.
T
e
knolog
i Ele
k
tro. 2004; 3(2)
.
[14]
Darma P
u
tra. Sistem Verifik
a
si Men
ggu
nak
an Ga
ris-
garis
T
e
lapak T
anga
n. T
e
knologi
el
ektro. 200
7;
6(2): 46-5
1
.
80
85
90
95
100
105
DB
15.1
D
B
15.2
D
B
15.3
D
B
25.1
D
B
25.2
D
B
50
64x64
pixels
128x128
pixels
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 11, No. 4, Dece
mb
er 201
3: 749
– 758
758
[15]
Don
ggi
ng Z
,
Y
ubi
ng
H, Ve
u
y
u T
.
Nonli
near
/N
on-Gauss
i
an
T
i
me Series
Predicti
on B
a
s
ed
on
RBF
-
HMM-GMM Model.
T
E
LKOMNIKA Indo
nesi
an Jo
urn
a
l
of Electrical
Eng
i
neer
ing.
201
2;10(6): 1
2
1
4
-
122
6.
[16]
Darma Putra.
Sistem Biometr
i
ka
. Yogy
ak
arta: Andi Offset.
2009: 161-
163.
Evaluation Warning : The document was created with Spire.PDF for Python.