TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 15, No. 2, August 201
5, pp. 294 ~
300
DOI: 10.115
9
1
/telkomni
ka.
v
15i2.823
9
294
Re
cei
v
ed Ma
y 8, 2015; Re
vised June
3
0
, 2015; Acce
pted Jul
y
15,
2015
Detection of Kidney Condition Using Hidden Markov
Models Based on Singular Value Decomposition
Siska Anra
e
n
i*, Ingrid Nurtanio, Indr
aba
y
u
Artificial Intel
lig
ent and Mu
ltim
edi
a Processi
n
g
Lab
orator
ium
,
Hasanu
dd
in
Univers
i
t
y
,
Perintis Kem
e
r
deka
an Km.10,
Makassar, Sul
a
w
e
si S
e
lata
n
902
45, Indo
ne
sia, 041
1-5
862
00
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: siska.anra
eni
@iee
e.org
A
b
st
r
a
ct
T
he frequ
enc
ie
s of chron
i
c ki
d
ney d
i
se
ase ar
e lik
ely to c
onti
nue to
incr
eas
e w
o
rldw
ide. S
o
pe
op
l
e
nee
d to tak
e
a
preca
u
tion, w
h
i
c
h is by
mai
n
ta
inin
g ki
dn
ey h
e
a
lth a
nd
ear
ly
detectio
n
of r
e
nal
i
m
pa
ir
ment
by
ana
ly
z
i
n
g
the
compos
ition
of
the iris
is k
n
o
w
n as
iri
dol
ogy
. T
h
is pa
per
pr
esents
a n
o
vel
appr
oac
h us
in
ga
one-
di
me
nsio
n
a
l discr
ete H
i
dde
n Markov
Mode
l (HM
M) classifier
and c
oefficie
n
ts Singu
lar V
a
lu
e
Deco
mpositi
o
n
(SVD) as a f
eature for
i
m
a
ge rec
ogn
iti
o
n
iris to in
dicat
e
nor
mal or
a
bnor
mal ki
dne
y. T
o
accel
e
rate a
l
g
o
rith
ms an
d reduc
e co
mp
utation
a
l
co
mple
xity and
me
mory consu
m
pti
on in
hardw
ar
e
imple
m
entati
o
n
s
, w
e
used in
a nu
mber of S
V
D small
c
oefficients a
nd
7-s
t
ate HMM for the i
m
a
ge
of the
mo
de
l confi
gur
ation.The syst
em
has
be
en e
x
amin
ed o
n
2
0
0
iris i
m
ag
es.The total
i
m
a
g
e
s
of the a
bnor
ma
l
kidn
ey con
d
itio
n w
e
re 100 i
m
ages a
nd thos
e for the nor
ma
l kidn
ey con
d
iti
on w
e
re 10
0 i
m
a
ges. T
he sy
ste
m
showed a classification rate up
to 100% using total of im
age fo
r training and te
sting the syst
em
unsp
e
cifie
d
, re
si
z
e
iris i
m
ag
e 56x4
6
pix
e
ls, c
oefficie
n
ts
of singu
lar va
lues
consists of ort
hog
on
al
matrix
is
(1,1) a
n
d
di
ag
ona
l
matric
es
are (
1
,1)
and
(
2
,2), qu
ant
i
z
e
d
val
ues [
18
10
7], a
nd c
l
assif
y
by
7-state
H
M
M
w
i
th .pgm for
m
at datab
ase.
Ke
y
w
ords
: irid
olo
g
y, SVD, HMM
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
One
part
of the hu
man
bo
dies th
at a
r
e
uniqu
e
an
d
can be
u
s
ed
a
s
the
syste
m
identifier
is iris of the human eye. The co
nditio
n
of the
orga
n or the level of a person'
s health can be
kno
w
n throug
h the iris ba
sed iridolo
g
y. Iridolo
g
y
is a sci
en
ce that studie
s
sig
n
s
contai
ned in the
netwo
rk
structure of the iri
s
a
s
a reflecti
on
of the con
d
ition and va
riou
s o
r
gan
s
and
system
s
in
the body.
The frequ
en
cies
of
chroni
c kid
ney di
se
a
s
e
are
li
kely t
o
continue to
incr
ease
w
o
rldw
ide.
In re
spon
se t
o
this, PERNEFRI (So
c
iet
y
of Neph
rolo
gy Indone
sia
)
appe
aled to t
he pu
blic to t
a
ke
pre
c
autio
na
ry actio
n
by
m
a
intaining
kid
ney he
al
th a
nd e
a
rly d
e
te
ction
of renal
impai
rment
[1].
Therefore,
we ch
ose to m
a
ke
a
syste
m
for early d
e
tection th
e ki
d
ney co
ndition
throug
h the i
r
is
image.
This p
ape
r prese
n
ts a n
o
vel app
roa
c
h
usin
g a on
e-dimen
s
ion
a
l discrete
HM
M (Hid
de
n
Markov Mo
de
l) a
s
a
cla
s
sifier a
nd
coefficients
SVD (Si
ngula
r
Valu
e
De
comp
ositio
n) a
s
a featu
r
e
for ima
ge
re
cognition
iri
s
.
We
used t
h
e
7-state
HM
M for the im
a
ge of
the m
o
del
config
urat
ion.
To a
c
celerate algo
rithm
s
and
red
u
ce
comp
utationa
l com
p
lexity and m
e
mo
ry co
nsu
m
ptio
n in
hard
w
a
r
e im
plementatio
n
s
, image
format se
gment
ation ki
dney
area
on the
iris im
age i
n
.jpeg
with different
sizes
ch
ang
e
d
format a
nd
size be
com
e
.pgm
and 56
x46pixels
whi
c
h i
s
then
used
in a num
be
r
of SVD small
coeffici
ents.
Re
sults
of
cal
c
ulatio
n of SVD co
efficien
ts are
then u
s
e
d
as
input to the HMM.
Previou
s
re
search o
n
det
ection of
kid
ney
con
d
ition
has be
en d
one. The
cla
ssifi
cation
use
s
ed
ge d
e
tection a
nd
segm
entation
floating wi
th
20 image
s iris teste
d
re
ach
ed 95% [2].
Re
sea
r
ch
u
s
i
ng Neu
r
al Networks Lea
rning
Ve
ct
o
r
Quanti
z
ation
ha
s tr
aining
accu
ra
cy10
0%
andte
s
ting a
c
curacy 93.7
5
%
[3].
Previou
s
re
search o
n
im
age recogniti
on with SV
D method a
n
d
HMM al
rea
d
y been
done. Retina
recognitio
n
, d
e
sig
n
of bloo
d dise
as
e re
cognition, an
d
ABO blood types
re
cog
n
ition
usin
g HMM
method
have
an a
c
curacy
rate of u
p
to
100% [4-6]. Face
re
co
gniti
on sy
stem u
s
ing
HMM a
c
hiev
e re
cognitio
n
accura
cy ra
te of
84.28%, with a databa
se of 7
0
image
s of 10
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction of Kidney
Con
d
i
t
ion Usi
ng Hi
dden Ma
rko
v
Model
s Base
d on… (Si
s
ka
Anraeni
)
295
individual
s
wi
th each indivi
dual h
a
ving 7
different
va
riations
of exp
r
essio
n
[7]. F
a
ce
re
co
gnition
usin
g SVM and HMM met
hod ha
s an a
c
cura
cy rate
of
97.78% for the ratio
of 50:50 trainin
g
data
image an
d 100% for the ratio 60:40 training data
i
m
age [8]. Therefo
r
we de
sign a sy
ste
m
to
detect the
kid
ney co
ndition
throug
h iri
s
i
m
age u
s
in
g
HMM b
a
sed
on SVD a
s
fe
ature
s
extra
c
t
i
on
method.
Whe
r
e the SVD wil
l
be use
d
to extract ir
is im
a
ge feature
s
a
nd HMM
will be used for iris
image
classifi
cation
sho
w
e
d
kidn
eys in
norm
a
l or a
b
norm
a
l co
ndit
i
ons. With th
e com
b
ining
of
SVD and HM
M, the detecti
on re
sults o
b
tained
will be more a
c
curate.
2. Rese
arch
Metho
d
2.1. Iridolog
y
Iridology
is the
sci
en
ce
o
f
analyzi
n
g
si
gns such
as colo
r,
a
nd structure of
the
iri
s
to
obtain imp
o
rt
ant informati
on abo
ut the
state of per
son’
s he
alth. Iris is a
circular di
sc-sha
ped
tissu
e locate
d in front of the lens. Iri
s
ha
s
sp
ecifi
c
advanta
g
e
s
, whi
c
h can
record all
state
orga
ns, bo
dy con
s
tru
c
tion,
as well as p
sycholo
g
ical
condition. Trace recording
s
relating to levels
of inten
s
ity o
r
d
e
viation
organ
s
cau
s
e
d
by di
se
ase reco
rde
d
i
n
a
system
atic a
nd
p
a
ttern
ed on
surro
undi
ng a
r
ea of iri
s
. It can be u
s
ed a
s
a pra
c
ti
cal guide to dia
g
nosi
s
of vario
u
s di
sea
s
e
s
.
The ki
dney is one of organ
that is locate
d in
both iri
s
e
s
. Iris ma
p is
divided into secto
r
s
and ea
ch
se
ctor is a
s
soci
a
t
ed with ce
rta
i
n body pa
rt
s. Kidney area
in the iris of the left eye is
locate
d at 6 and 7 o'
clo
c
k, while the rig
h
t eye ir
isis l
o
cate
d at 5 and 6 o’clo
c
k. Figure1 shows a
map of the rig
h
t and left eye by Dr. Bern
ard Jen
s
en,
D.C.
Figure 1. Iridology Ch
art d
e
velope
d
by Dr. Bernard Jensen, D.C. [9]
2.2.
Data Sets a
n
d Image Pre-proces
sing
For thi
s
study
, 200 iri
s
ima
ges
co
nsi
s
t o
f
kidney cond
ition types (a
bnormal an
d
norm
a
l)
wa
s prepa
re
d from 20 p
e
r
so
ns. Th
ere are 1
0
image
s pe
r perso
n. So the
imag
e dataset of the
abno
rmal
kid
ney co
ndition
is 1
00 im
ag
es a
nd fo
r th
e no
rmal
kid
ney co
ndition
is 1
00 im
ag
es.
Every perso
n
we
re ta
ken
a
t
different tim
e
s, vary
in
g th
e lighting, ey
es exp
r
e
s
sio
n
s
(eye
s op
e
n
e
d
perfe
ctly or
sl
ightly clo
s
ed
), and p
upil im
age d
e
tail (sh
r
ink an
d swel
l due to
ca
m
e
ra fla
s
h li
ght
).
The imag
es a
r
e in
.jpeg
format. The size
of each imag
e is 259
2x19
44 pixels.
For the fi
rst
step, image
d
a
taset i
s
divi
ded in t
w
o p
a
rts
per
pe
rson. So the to
tal image
for trainin
g
is 100 imag
es
and for te
stin
g are 1
00 im
age
s. Next st
ep, SVD is u
s
ed to extra
c
t
feature
s
fro
m
iris im
age
s a
nd HM
M is
used to buil
d
th
e cla
s
sificatio
n
model.
HM
M is trai
ned
with
extracted fea
t
ures from five iris imag
es and test
ed
with five iris image
s that different with the
image
s used
in training. Th
e model returns proba
b
ilities of ho
w like
l
y the iris ima
ges recogni
zed
as ab
normal
kidn
ey con
d
ition
and n
o
rm
al kidn
ey con
d
ition.
This
system
usin
g the dat
a input form
the colo
ur iri
s
image
RG
B transfo
rme
d
into a
gray-scale
im
age [1
0]. Before it
ca
n b
e
use
d
for
furth
e
r im
age
pro
c
e
ssi
ng, p
a
rt
of the iri
s
im
a
g
e
on all part
s
of the eye is separated thro
ugh the firs
t i
r
is ima
ge localizatio
n pro
c
ess. Re
sults
of
the iri
s
image segmentation still have a
low level
of
contrast and iris
fibres im
age detail i
s
less
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 294 –
300
296
clea
r that produ
ce lo
w a
c
cura
cy. The
r
efore, the image of th
e
iris contra
st
enhan
ce
d u
s
ing
Adaptive Hist
ogra
m
Equali
z
ation [11]. T
h
is metho
d
ai
ms to obtain i
m
age
s with g
ood contra
st but
will n
o
t da
m
age th
e
overall im
age
q
uality. The i
m
age
then
resi
zed
to
56
x46 pixel
s
a
s
rep
r
e
s
ente
d
on Fig
u
re
2(c). In o
r
d
e
r
to balan
ce
the flash effe
ct and
re
du
ce the n
o
ise, a
minimum o
r
d
e
r n
online
a
r-static filter i
s
use
d
. The filt
er give
s
smo
o
thing effe
ct and
redu
ce
s
the
image noi
se.
An example result of the a
pp
lied filter is
pre
s
ente
d
on
Figure 2(d
)
.
2.3.
Selection of
Region o
f
Interes
t
(ROI)
ROI is th
e area u
s
ed to e
x
tract feature
s
. Fo
r thi
s
re
sea
r
ch, the
ROIs
are th
e
kidn
ey
orga
n on th
e
iris im
age
s.
All ROIs
are manu
ally
selecte
d
from
the image
b
y
a well-t
r
ain
ed
operator a
n
d
confirm
ed b
y
iridologi
st. A ROI of
many size pixel
s
wa
s extra
c
ted with a mass
centred in th
e wind
ow [12
]. Figure 2(a) and 2(b)
rep
r
esents th
e ROI of iris ima
ge to get kid
ney
orga
n. The f
l
owcha
r
t of kidn
ey org
a
n
conditio
n
cl
assificatio
n
u
s
ing
HMM i
s
presented
in
Figure 3.
Figure 2. (a)
The sel
e
cte
d
ROI of image
;
(b) The
ROI of Iris for Kid
ney Orga
n;
(
c) The resi
zed
iris ima
g
e
s
to 56x46 pixels;
(d) Th
e effect of the smoothing filter
Figure 3. Stage of the kidn
ey organ
con
d
ition cla
s
sification meth
o
d
2.4.
SVD for Fea
t
ure Extra
c
tio
n
and Quan
ti
zatio
n
SVD is a
too
l
often
use
d
in si
gnal
p
r
o
c
e
ssi
ng
and
statistical d
a
ta an
alysi
s
. S
i
ngula
r
values from
g
i
ven data
mat
r
ix co
ntain th
e inform
ation
about
noi
se l
e
vels, e
nergy, ran
k
of
matrix,
etc.
Sing
ular vectors of
ma
trix
are
the
span
ba
se
s of
matrix, an
d orthog
onally norm
a
l,
they can
sho
w
few fea
t
ures of the p
a
tterns in the
signal.
SVD gives a different way to extract algeb
raic
feature
s
from
an image ve
ctor. SVD fro
m
matrix
X
is defining by:
∗
Σ
∗
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction of Kidney
Con
d
i
t
ion Usi
ng Hi
dden Ma
rko
v
Model
s Base
d on… (Si
s
ka
Anraeni
)
297
Whe
r
e
U
and
V
are orth
ogon
al matri
c
e
s
and
Σ
is a diagon
al m
a
trix of singu
lar value
s
[13].
Coeffici
ents
U(1,1
)
,
(1,1) and
(2,2
)
are cho
s
e
n
by trial and error a
s
the most rel
e
vant feature
s
of the image [8]. Each block is represent
ed by
a vecto
r
of observati
on with n ele
m
ents:
,
,…
,
(2)
Every eleme
n
t of (2) is quantized into
D
i
distin
ct levels. The differen
c
e bet
wee
n
two quanti
z
ed
values i
s
:
(3)
Whe
r
e
coeff
im
a
x
and
coeff
im
i
n
are maximum and mi
nimum of the
coefficie
n
ts i
n
all obse
r
va
tion
vectors. Every element of vector
C
is re
placed with it
s value that h
a
s be
en qu
an
tized:
(4)
Dist
in
ct
(
D
i
) value
s
used in
(4) to qu
anti
z
e coefficie
n
ts
U(1,1),
(1,1
),
(2,2)
a
r
e 1
8
, 10, and 7.
These value
s
are
cho
s
e
n
based o
n
the
experim
enta
l
results in [8
]. Next step i
s
to re
present
each blo
ck by
one discrete value name
d
label
:
∗
1
0∗
7
∗7
1
(5)
Whe
r
e
qt
1
,
qt
2
and
qt
3
are
the quanti
z
e
d
values. If the co
efficient
s (2
) are all
zero, the
label
value is one
and if quantized the first feature
U(1,1)
in
to 18 levels, the se
con
d
feature
(1,1)
into
10, and the third featu
r
e
(2
,2)
into 7, lea
v
ing 1260 p
o
ssi
ble di
stinct
vectors.So the
label
will be
given the m
a
ximum value
1260. A
s
the
result, ea
ch
iris ima
ge i
s
re
pre
s
ente
d
by
an ob
se
rvatio
n
seq
uen
ce
s
with 52 or
60 o
b
se
rved
state
s
, equival
ent
to blocks nu
mber. T
he 7
-
state HMM m
odel
use the
s
e o
b
s
ervatio
n
vectors a
s
input.
2.5.
HMM fo
r Trai
ning and Cla
ssifica
tion
HMM i
s
modeling the probability of
a
system
to look for paramet
e
rs Markov
(hidden),
whi
c
h i
s
not
known to
a
c
qu
ire th
e
syste
m
s
analy
s
is.
HMM i
s
abl
e t
o
ha
ndle
the
cha
nge
stati
s
tics
of the image by modeling t
he eleme
n
ts
usin
g pro
babi
lities. HMM
s
are u
s
ually u
s
ed to mod
e
l one
dimen
s
ion
a
l data. HMM a
l
so be
used for tempo
r
al
p
a
ttern cl
assifi
cation
syste
m
. For exam
ple,
voice
re
cog
n
i
tion, han
dwriting, ge
sture
s
, bio
in
f
o
r
m
at
ics,
se
nt
en
ce com
p
r
e
s
s
i
on,
et
c.
E
v
e
r
y
HMM i
s
a
s
so
ciated
with hi
dden
state
s
a
nd ob
se
rvabl
e se
que
nce g
enerated
by t
he hid
den
sta
t
es
individually.
Cla
ssifi
cation
pro
c
e
ss i
s
b
a
se
d on front
al iris vie
w
.
From top to
bottom the iri
s
imag
e
can
be
divided into
seve
n re
gion
s
wh
ich e
a
ch i
s
assign
ed to
a state i
n
a
left to right
one
dimen
s
ion
a
l HMM. Figu
re
4 sho
w
the
seven ir
i
s
region
s. Figu
re 5 sho
w
s equivalent o
ne-
dimen
s
ion
a
l HMM mod
e
l for a divided i
m
age into se
ven distin
ct region
s a
s
se
en in Figu
re 4
.
Figure 4. Iris
regio
n
s from top to bottom
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 294 –
300
298
Figure 5. A one dimen
s
io
n
a
l HMM mod
e
l with 7-states for a
n
iris i
m
age by sev
en regi
on
s
The funda
me
ntal factors to
build a HMM
model are as follows:
a)
N=|S| is the
number of
st
ates in HMM
model,
where
,
,…,
is
nu
mb
er
o
f
all possi
ble st
ates. The state model at a
n
y random time t is given
by
∈
,
1
where
T
is th
e observation
sequ
en
ce le
ngths.
b)
M=
|V| is
the number of
s
y
mbols
of
different vec
t
or ob
s
e
rvations
, where
,
,…,
is nu
mbe
r
of
all po
ssi
ble
vector fe
ature
. The featu
r
e vecto
r
at a
n
y
rand
om time t is given by
∈
.
c)
‘A’ is the probability of state
transition m
a
trix which giv
en by
,
where
,
|
,
1
,
,
0
1
(6)
∑
1
,
1
(7)
So the transiti
on matrix
A
as sh
own belo
w
on Tabl
e 1.
Table 1. Tran
sition Matrix
A-7 state[14]
1 2 3 4 5 6
7
1
0,5
0,5
0 0 0 0
0
2
0
0,5
0,5
0 0 0
0
3
0 0
0,5
0,5
0 0
0
4
0 0 0
0,5
0,5
0
0
5
0 0 0 0
0,5
0,5
0
6
0 0 0 0 0
0,5
0,5
7
0 0 0 0 0 0
1
d)
‘B’ is the probability of the
observa
tional
vector matrix
which given
by
,
w
h
er
e
|
,
1
,
1
(8)
And the emission mat
r
ix
B
is
:
1
2 … 1260
11
11
…1
…1
⋮⋮
11
⋱⋮
…1
/
1260
e)
′
′
, is the distrib
u
tion of the initial state,i.e
w
h
er
e
,
1
(9)
Assu
ming th
a
t
a block mov
e
s fro
m
top t
o
bottom
of the iri
s
imag
e
s
an
d in any t
i
me that
block
sho
w
s
one of the
seven re
gion
s.
The bl
oc
k i
s
shifting
con
s
eque
ntly and
can
not mi
ss a
state. For ex
ample, a
blo
c
k in the state
“s
2
”, the n
e
xt state ca
nnot
be “s
1
”, but will always at t
he
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Dete
ction of Kidney
Con
d
i
t
ion Usi
ng Hi
dden Ma
rko
v
Model
s Base
d on… (Si
s
ka
Anraeni
)
299
state “s
3
”. Th
us the proba
bility of movi
ng from on
e st
ate to the next state is 50% and rem
a
in in
the current
st
ate is 50%. I
n
itia
l state
of
the
system i
s
“s
1
” with a
probability of
1. And the fi
nal
state of the system is alwa
ys “s
7
”. So the
matrix is
:
1
0
0
0
0
0
0
In short n
o
tation HMM i
s
d
e
fined a
s
follows:
,
,
(10
)
,
matrices d
e
fine the iris m
o
del t
hat is trai
ned with the
dataset.
3. Results a
nd Analy
s
is
3.1.
Comparing
Resul
t
s fo
r Differen
t Total and Size o
f
Training Image
The
cla
s
sif
i
ca
t
i
on sy
st
em
w
a
s t
e
st
ed
o
n
a ma
chi
ne
wi
th Co
re
i3
CP
U 1.8
0
GH
z,3
.
89 GB
Ram
and
6
4
-bit ope
rating
system.
The
bestresul
tsa
r
e sho
w
n
on
Table2.Ba
se
d on
the
tota
l of
image for trai
ning and te
sting, image wi
th size
d 56x
46 pixels giv
e
s better
cla
ssifi
cation re
sult
rate
up to
10
0%. The i
n
tui
t
ion be
hindthi
s
re
sult i
s
th
at total of im
age fo
r traini
ng a
nd
sm
all iri
s
image detail
s
are not impo
rtantand m
a
y even wo
rsen
the cla
ssifi
cat
i
on.
Table 2. Co
m
parin
g Re
sult
s for Different
Total and Size of Trainin
g
Image
Total
of I
m
a
g
e
Classific
a
tio
n
R
a
te
For Trai
nin
g
For Te
stin
g
For 56x4
6 pixel
s
For 64x6
4 pixel
s
100 100
100%
15%
120 80
100%
15%
140 60
100%
15%
160 40
100%
15%
180 20
100%
15%
3.2. Classific
a
tio
n
s
Ra
te
Fo
r Differen
t Fea
t
ures
Selection
an
d arrang
eme
n
t of the SVD featur
es i
s
cru
c
ial fo
r the pe
rform
a
n
c
e of the
HMM syste
m
as it’s prese
n
ted on Tabl
e 3. Coefficie
n
ts
U(1,1),
(1
,1) and
(2,2
)
are chosen
by
trial and e
r
ror as the mo
st relevant featu
r
es from
ima
ges a
nd give
s better
cla
s
sification rat
e
up
to 100%.
Table 3. Cla
s
sificatio
n
s
Ra
tefor Differe
nt Feature
s
Used Val
u
es
Classific
a
tio
n
Rate
1st
2nd
3rd
U(1,1)
Σ
(1,1)
Σ
(2,2)
100%
Σ
(3,3)
Σ
(1,1)
Σ
(2,2)
20%
U(1,1)
Σ
(1,1)
V(1,1)
10%
U(2,2)
Σ
(1,1)
Σ
(2,2)
60%
U(1,1)
Σ
(1,1)
V(2,2)
5%
Σ
(1,1)
Σ
(2,2)
Σ
(3,3)
5%
3.3.
Classific
a
tio
n
Rate For Differe
nt Qu
a
n
tize
d Value
s
The qua
ntiza
t
ion levels 18
, 10 and 7 h
e
lps ma
ke th
e algorith
m
faster, b
u
t with lowe
r
image resolut
i
on. With that
such qua
ntization levels
it will be difficul
t
or impossibl
e to classify iris
image
s with
bad qu
ality.On Table 4 we
can see that
the quantiza
t
ion levels [1
8 10 7], [7 7
7
]
and [1
8 18
1
8
] gives
bette
r cl
assificatio
n
rate
up
to
100%.Whil
e
quanti
z
ation l
e
vels [3
2 32
32]
and [64 64
6
4
] occurs co
mputation p
r
oblem be
ca
u
s
e tho
s
e valu
es can’t usi
n
g in HMM tra
i
ning
pro
c
e
ss
whe
r
e value of se
quen
ce mu
st
con
s
i
s
t of integers between
1 and 126
0.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 294 –
300
300
Table 4. Cla
s
sificatio
n
s
Ra
tefor Differe
nt Quantized Value
s
Qua
n
tized V
a
lu
es
Possible
Com
b
ina
t
io
ns
Classific
a
tio
n
R
a
te
1st
2nd
3rd
18 10
7
18*10*7=1260
100%
7 7
7
7*7*7=343
100%
18 18
18
18*18*18=5832
100%
32 32
32
computation
pro
b
lem
64 64
64
computation
pro
b
lem
4. Conclusio
n
Dete
ction
of
kidn
ey condit
i
on throug
h t
he ir
i
s
i
m
age
ca
n b
e
d
o
n
e
by u
s
in
g S
V
D a
s
a
feature extra
c
tion an
d HM
M as a cl
assi
fier. The
exp
e
rime
nts sho
w
ed a
cla
ssifi
cation rate u
p
to
100%
by u
s
i
ng total
of im
age fo
r t
r
aini
ng a
nd te
st
in
g the
sy
stem
un
spe
c
ified,
resi
ze
iri
s
i
m
age
56x46 pixel
s
, coeffici
ent value
s
U(1,1),
Σ
(1,1)
and
Σ
(2,2)
, qua
ntized value
s
[1
8 10 7], and
cla
ssify by 7-state HMM
with
.pgm
format databa
se. These cl
assi
f
i
cation
results are obtain
e
d
usin
g
cho
s
e
n
optimized
system p
a
ra
m
e
ters after
a
co
mprehe
nsive study
ab
out their affe
ct
discu
s
sed in this pa
per.
Referen
ces
[1]
W
i
ka.
T
he Virtue of Dri
n
kin
g
Enou
gh W
a
te
r for Health K
i
dne
y.
Jo
urna
l
of Medica
l On
line.
20
13
;
39(0
6
).
[2]
Rah
a
y
u D
N
P.
Appl
icati
ons
of Kidn
e
y
D
i
s
o
rder
s
Dia
gn
o
s
is T
h
rough Ir
is E
y
e Us
in
g
Segme
n
tatio
n
Method Bas
ed
Edge D
e
tectio
n.
T
hesis. Semaran
g
: Dipo
n
e
goro U
n
ivers
i
t
y
. 2013.
[3]
Putra AP, Sut
o
jo
T
.
Identification
of Ki
dn
e
y
F
u
nctio
n
D
e
crease
Co
nditi
ons T
h
roug
h I
r
is E
y
e
Usi
n
g
Neur
al Net
w
o
r
k Learn
i
ng V
e
c
t
or Quantizatio
n Method
.T
ec
h
no.COM
. 2014;
13(1): 45-5
2
.
[4]
Yuli
anti SM. Retina R
e
co
gniti
on Usi
ng Hi
dd
en Markov Mo
del. T
hesis. Indones
ia Un
ivers
i
t
y
. 20
08.
[5]
Lestari AP. D
e
sig
n
of Blo
o
d
Diseas
e Rec
ogn
iti
on
Usin
g
Hidd
en M
a
rkov Mod
e
l. T
hesis. Indon
es
i
a
Univers
i
t
y
. 2
0
0
8
.
[6]
Sasmita C. Recog
n
itio
n of ABO Blood T
y
p
e
s Usin
g H
i
dd
en Markov
Model. T
hesis. Indonesi
a
Univers
i
t
y
. 2
0
0
8
.
[7]
Seprita
hara.
F
a
ce
Reco
gn
iti
on S
y
stem
Us
in
g
Hi
dd
en M
a
rkov M
ode
l
Methods. T
h
e
s
is. Indo
nes
i
a
Univers
i
t
y
. 2
0
1
2
.
[8]
Naimi HM, Dav
a
ri P. A Ne
w
Fast and Efficie
n
t H
MM-Based
Face Recog
n
it
ion S
y
stem Usi
ng a 7-Stat
e
HMM Alon
g W
i
th SVD Coefficients.
Iranian Jour
nal
of Electrical &
Electrical Engineering (IJEEE).
200
8; 4(1): 46-
57.
[9] https://detox
.net.au/iridology
-chart/
[10]
Anrae
n
i S, Ind
r
aba
yu, N
u
rtan
io I.
F
eature E
x
traction of Iris
Imag
e Usi
ng
Grey Level
Co
-occura
nce
Matrices.
N
a
ti
ona
l C
onfer
en
ce o
n
Comp
uter Sci
ence
En
gin
eeri
n
g
(SN
R
IK). Moslem
Un
iversit
y
o
f
Indon
esi
a
Mak
a
ssar. 201
5; 1.
[11]
Ilham AA, In
drab
a
y
u, H
a
s
anu
ddi
n R, P
u
tri DM.
Wavelet Analysis for Id
e
n
tifica
tion of
Lu
n
g
Abnor
maliti
e
s
Using
Artificia
l
Neur
al N
e
tw
ork.
Makassar
Internatio
na
l
Confer
enc
e
on El
ectrical
Engi
neer
in
g an
d Informatics (MICEEI). 2014
: 156-16
0.
[12]
Nurtanio I, Ast
u
ti ER, Pur
nama
IKE, H
a
ri
ad
i
M, Pu
rn
o
m
o
MH
. Cl
a
s
si
fy
i
n
g Cy
st an
d
T
u
mo
r L
e
s
i
o
n
Using
Sup
port
Vector Mach
i
ne Bas
ed
on
D
enta
l
Pa
nor
amic Imag
es
T
e
xture F
eatu
r
es.
IAENG
Internatio
na
l Journ
a
l of
Co
mputer Scie
nce (
I
JCS).
2013; 4
0
(1).
[13]
Sadek
RA. SV
D Base
d Imag
e Process
i
n
g
A
pplic
at
io
ns: State of T
he Art, Contri
butio
ns a
nd R
e
se
arc
h
Chal
le
nges.
Int
e
rnati
ona
l Jo
ur
nal
of Adv
anc
ed C
o
mp
ut
er
Scienc
e a
nd A
pplic
atio
ns (IJ
A
CSA).
20
12;
3(7): 26-3
4
.
[14]
Dinkov
a
P, Ge
orgi
eva P, Mi
l
anov
a M.
F
a
c
e
Rec
o
g
n
itio
n
Using
Sin
g
u
l
ar
Valu
e D
e
co
mpositi
on
and
Hidd
en M
a
rkov
Mode
ls.
T
he 16th Intern
ation
a
l C
onfere
n
ce
on
Math
ematic
al Meth
ods, C
o
mputati
o
n
a
l
T
e
chniques
an
d Intelli
ge
nt S
y
stems (MAMECT
IS '
14). 2014: 144-1
49.
Evaluation Warning : The document was created with Spire.PDF for Python.