TELKOM
NIKA
, Vol.14, No
.3, Septembe
r 2016, pp. 1
128
~11
3
3
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v14i3.3486
1128
Re
cei
v
ed Fe
brua
ry 15, 20
16; Re
vised
May 4, 201
6; Acce
pted Ma
y 27, 201
6
Classification of Motorcyclists not Wear Helmet on
Digital Image with Backpropagation Neural Network
Sutikno*, Ind
r
a Was
p
ada,
Nurdin Ba
htiar, Priy
o
Sid
i
k Sasongko
Dep
a
rtment of Comp
uter Scie
nce/Informati
cs
, F
a
cult
y
of Sci
ence a
nd Math
ematics,
Univers
i
t
y
of
Di
pon
eg
oro
Prof. Soedarto
Street,
T
e
mbal
ang, Semar
a
n
g
502
75, Indo
n
e
sia
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: tik@und
ip.ac
.
id
A
b
st
r
a
ct
One of t
he w
o
rld
’
s
l
ead
in
g
caus
es
of d
eath
is
traffic
acci
dents.
D
a
ta fro
m
W
o
r
l
d H
ealth
Organi
z
a
t
i
on (
W
HO) that
there are 1.
25
mil
lion p
e
o
p
le i
n
the w
o
rld di
e each year. Mea
n
w
h
ile, bas
ed
o
n
data o
b
tai
ned
from Statistics
Indon
esia, tra
ffic acci
de
nts from
200
6 to 2
013 c
ontin
ue t
o
incre
a
se. Of al
l
these accidents,
the largest accident occur
r
ed at motorcy
c
lis
ts, especially m
o
torcyclists who not wearing
standar
d he
lmet. In controlli
ng the
motorc
yclists, polic
e
view
directly a
t
the hi
ghw
ay, so that there ar
e
w
eaknesses
w
h
ich th
ere
are
still a
possi
bi
lit
y of motor
cycli
s
t offenders w
ho ar
e u
n
d
e
tec
t
able
esp
e
cia
l
l
y
for
motorcyc
lists
w
ho are n
o
t w
ear h
e
l
m
et. T
h
is pa
per ex
pl
ai
ns rese
arch o
n
i
m
a
ge cl
assi
fication
of hu
man
hea
d w
earin
g
a hel
met an
d n
o
t w
earing
a h
e
l
m
et w
i
th
bac
kprop
agati
on n
eura
l
netw
o
rk alg
o
rith
m.
T
he test
results of this analysis is th
e app
licati
on
can perfo
r
m
s
classificati
on
w
i
th 86.67% a
ccuracy rate. T
h
is
researc
h
can be dev
eloped into a larger system
and
integrated that can be used
to detect m
o
torcyclis
ts
w
ho are not w
earin
g hel
met.
Ke
y
w
ords
:
tra
ffic accidents, classificati
on, n
o
t w
ear
ing a h
e
l
m
et, backpr
o
pag
atio
n neur
a
l
netw
o
rk
Copy
right
©
2016 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Traffic a
cci
de
nts are o
ne cause of deat
h in t
he worl
d
.
Data obtain
ed from Worl
d Health
Orga
nization
(WHO
), the n
u
mbe
r
of pe
ople who di
e
d
in the worl
d ca
use
d
by
traffic accid
e
nts
totaled 1.25
million pe
opl
e ann
ually [1].
While in I
ndon
esi
a
, ba
sed
on d
a
ta
from Statisti
cs
Indone
sia, nu
mber of a
c
cid
ents record
e
d
from 20
06 t
o
2013
co
ntin
ues to in
crea
se.
Of all the
s
e
accide
nts, th
e big
g
e
s
t a
c
cide
nt o
c
curred o
n
m
o
torcycli
sts [2].
One
of the
reason
s that
the
motorcycli
st were not we
a
r
ing a hel
met acco
rding to
stand
ard
s
set by the government.
The g
o
vernm
ent ha
s i
s
sue
d
re
gulatio
n
No. 2
2
of
200
9 on
traffic
a
nd road
tran
sportation
whi
c
h
one
of
the pu
rpo
s
e
i
s
to
de
crea
se the
deat
h
rate that
cau
s
ed by t
r
affic
accide
nt. On
e of
the content
s i
n
the
re
gulati
on that
i
s
in
a
r
ticle
106
p
a
ragra
ph
7
stat
es th
at eve
r
y
person
d
r
ivin
g a
motorcycle
a
nd the moto
rcycle
pa
ssen
ger m
u
st
wea
r
a hel
met th
at meets n
a
tional
standa
rd
s of
Indone
sia [3].
Mean
while, a
ll this time, supervi
sion of
motorc
y
c
list
s
on the roa
d
by the police is
still
done
man
uall
y
by loo
k
ing
dire
ctly at th
e hig
h
way. There
a
r
e
we
a
k
ne
sse
s
whi
c
h the
r
e i
s
still
a
possibility of
motorcycli
st off ende
rs
who
are
und
etectabl
e, especi
a
lly for
motorcycli
sts wh
o
were not we
a
r
ing hel
mets,
still if the police is
st
andi
n
g
guard at the police
stati
on, they can
no
t
kno
w
m
o
torcyclists
wh
o were
not weari
ng helm
e
ts.
One
solutio
n
to overcome t
hat problem i
s
to
utilize the camera in
conducti
ng surveillance on highways a
nd combined with the detection
pro
c
e
ss fo
r d
e
tecting the p
r
esen
ce of m
o
to
rcy
c
list
s
who we
re not
wea
r
ing h
e
lm
ets.
There are se
veral rese
arches th
at have been
don
e
relating to t
he monito
rin
g
of the
traffic that i
s
usin
g a
came
ra
with
good
result
s [4
-6].
On tho
s
e
researche
s
, the
came
ra
was
only
use
d
for mon
i
toring
and
im
age from th
e
result of
re
co
rding
was not
used
well. T
he result of t
he
recording
ca
n
be u
s
e
d
for
example to
d
e
tect o
r
to
cl
assify motorcyclist wea
r
ing
a helm
e
t or
not
automatically. Resea
r
ch from this cl
assificati
on issu
e
s
ha
s bee
n d
one by usi
n
g
Support Ve
ctor
Machi
n
e
s
(S
VM) to gene
rate an accu
ra
cy rate of 85
% [7].
This pap
er e
x
plains
th
e classificatio
n
p
r
oc
ess of
mot
o
rcyli
s
ts we
aring
a helmet and
n
o
t
on the
high
way on di
gital
image
with b
a
ckpropa
gati
on ne
ural net
work. T
h
is
m
e
thod h
a
s be
en
widely u
s
e
d
in the p
r
o
c
e
ss
of identifi
c
ati
on and d
e
tection with
good re
sult
s,
for
exa
m
p
l
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cla
ssifi
cation
of Motorcycli
s
ts not Wea
r
Helm
et on Di
gital Im
age With… (Sutikn
o
)
1129
identificatio
n
of varietie
s of
f
ood,
stone t
e
xture id
entification,
s
hap
e
identificatio
n
,
moldy pe
an
ut
kernel
s id
enti
f
ication, ide
n
tification of
external
quality
of wh
eat g
r
ain, de
rmatol
ogical di
sea
s
es
detectio
n
,
an
d re
nal tum
o
r detectio
n
[8-18]. The
re
su
lt
s of
t
h
is
re
s
ear
ch
m
a
y contribute
to the
developm
ent of
violation d
e
tectio
n
sy
st
e
m
s,
of
mot
o
r
c
y
c
list
s
not
we
aring
a
helm
e
t, automatical
ly
based on di
gi
tal image that captured by the cam
e
ra.
2.
Res
earc
h
Method
2.1. Sy
stem
Des
c
ription
De
scription of
application that
unde
rtake
n
in this re
se
arch ca
n be seen in Figu
re
1.
Figure 1. De
scriptio
n of Applicatio
n
Figure 2. Architecture of Backpropa
gati
on Ne
ural
Ne
twork
In the outlin
e, the syste
m
t
hat wa
s
built is divid
ed into two
parts, n
a
mel
y
training
p
r
oc
es
s
an
d
te
s
t
in
g
pr
oc
ess
.
a.
The traini
ng p
r
ocess
The traini
ng
pro
c
e
ss b
egi
ns by ente
r
in
g some
data
in the form of motorcy
c
li
st image
that wea
r
h
e
l
m
et and
not
wea
r
h
e
lmet.
After t
hat, the data i
s
carried
out a p
r
oce
s
s
of ima
ge
pro
c
e
ssi
ng
so the
re
sult
is im
age
feature.
Fu
rth
e
r, featu
r
e i
m
age
is sto
r
ed in
the
d
a
ta
base.
Feature imag
e
tha
t
has
bee
n sto
r
ed
was traine
d
usin
g
ba
ckprop
agatio
n
n
eural
netwo
rk.
Th
e
result
s of this trainin
g
are in t
he form of weighte
d
values
of backp
rop
agat
ion
neural network archite
c
ture
.
b.
The testing p
r
oce
ss
The te
sting p
r
ocess i
s
u
s
e
d
to dete
r
min
e
ac
cu
ra
cy l
e
vel of the
system that h
a
s b
een
made.
Thi
s
p
r
ocess i
s
sta
r
ted in the form of
input image.
Then
do
the image p
r
oce
s
sing
whi
c
h
will g
ene
rate
image fe
ature.
In testing
pro
c
e
ss,
dat
a inp
u
t is in
the fo
rm of
wei
ghting
value
resulting fro
m
the trainin
g
pro
c
e
s
s an
d the
feat
ure
image
in the
testing p
r
o
c
e
ss.
Th
e re
sult
s of
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 3, September 20
16 : 1128 – 11
33
1130
this process
are u
s
e
d
to test wheth
e
r
the
image i
s
an image of
the head of
a motorcycli
st
wea
r
ing a h
e
l
m
et or not.
Table 1. Example of Data
T
hat Will be
Trained and Tested
(a). T
he image of
head not
wear
a helmet
(b). T
he image of
head
w
e
a
r
helmet
F
ile Name
Image
F
ile Name
Image
File001
File016
File002
File017
File003
File018
File004
File019
File005
File020
File006
File021
File007
File022
File008
File023
File009
File024
File010
File025
File011
File026
File012
File027
File013
File028
File014
File029
File015
File030
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cla
ssifi
cation
of Motorcycli
s
ts not Wea
r
Helm
et on Di
gital Im
age With… (Sutikn
o
)
1131
2.1. Architec
ture o
f
Ba
ck
propag
a
tion
Neur
al Net
w
ork
Backpropa
ga
tion neu
ral n
e
twork a
r
chit
ecture of
this system
con
s
ists of 40
0 in
puts, on
e
hidde
n layer
con
s
i
s
ts of 40
neuro
n
s, an
d
one output a
s
in Figu
re 2.
3. Resul
t
s
and
Discus
s
ion
This
re
sea
r
ch
con
d
u
c
ted t
w
o te
sts, n
a
m
ely
ba
ckpro
pagatio
n ne
ural network tra
i
ning a
nd
perfo
rman
ce
of backpropa
gation ne
ural
netwo
rk al
go
rithm.
3.1.
The Tes
t
ing
of Ba
ckpr
o
p
a
gation
Neur
al Net
w
o
r
ks
Training
The data that
used a
s
train
i
ng amou
nted
to 150 images whi
c
h
con
s
ist
s
of 75 image
s of
the head we
aring a h
e
lm
et and 75 im
age
s of t
he head not we
aring a helmet.
The image size
that use
d
a
s
trainin
g
is
20x
20 pixel, the
example i
s
in
the Tabl
e 1.
The pu
rp
ose
of this trai
nin
g
is
to find the best suita
b
le
netwo
rk
pattern
s of
the architectu
re t
hat has b
e
e
n
created
which
prod
uces val
ues of net
work wei
ghts.
Testing
of net
work t
r
ainin
g
is u
s
ing
traini
ng inte
rface a
s
in
Figu
re 3
with same i
n
put limit
of epoch a
nd error
wh
ich maximu
m limit of
epo
ch is 3
0
.
000 and b
ound
ary error is
0.0000
01.
Thi
s
test
will fin
d
the influe
nce of variatio
n
s
in
rate
of le
arnin
g
(
α
) bet
wee
n
0.1 u
p
to
0.9, then sea
r
che
d
the lowest epo
ch. Th
ese te
st re
sul
t
s are
sho
w
n
in Table 2.
Figure 3. Trai
ning Interfa
c
e
of
Backp
rop
agation Neu
r
al
Network
Table 2. Te
sting Trai
ning
Result of Backp
rop
agatio
n Network with
A Learni
ng Rate (
α
) Variati
o
n
No.
Learning r
a
te (
α
)
Epoch
Error
1 0.1
5988
0.000001
2 0.2
3055
0.000001
3 0.3
2103
0.000001
4 0.4
1451
0.000001
5 0.5
1129
0.000001
6 0.6
933
0.000001
7 0.7
914
0.000001
8 0.8
1016
0.000001
9 0.9
1049
0.000001
From Ta
ble 2
it appears that the maximum lim
it of th
e same e
p
o
c
h and the sa
me error
make all
the testing
h
a
s reached a
pre
deter
mi
ned
e
rro
r befo
r
e
re
achi
ng the m
a
ximum ep
och
limit.
Additionally, the smal
lest epo
ch i
s
achi
eved
du
ri
ng the test b
y
giving learn
i
ng rate by 0.
7
with 9
14
epo
ch. M
o
re
over, it can
be
seen th
at t
he
large
r
th
e le
arnin
g
rate v
a
lue, the
fast
e
r
provisi
on give
s tende
ncy in
achievin
g an
erro
r or e
p
o
c
h value be
co
me small
e
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 14, No. 3, September 20
16 : 1128 – 11
33
1132
3.2. Testing
Performan
c
e
Backp
ropa
g
a
tion Neural
Net
w
o
r
k
Alg
o
rithm
This t
e
st i
s
u
s
ed
to me
asu
r
e p
e
rfo
r
ma
n
c
e
of
ba
ck
p
r
opag
ation ne
ural network algorith
m
in cla
s
sifying
the imag
e
of a hum
an
head
we
arin
g helmet
s
a
nd not
wea
r
i
ng a h
e
lmet.
The
interface u
s
e
d
in these te
sts a
s
in Fig
u
re 4.
Thi
s
test is u
s
ing
data from th
e netwo
rk weights
training result variation of learni
ng rate.
Input fr
om these te
sts are 30 image
s wi
th a size of 2
0
x
20 pixels co
mposed of 2 types of 15 image
s of
human hea
ds were not we
ari
ng helmet
s
and 15
image
s of m
an we
arin
g a
helmet as
shown in Tabl
e 1. The testi
ng re
sult
s pe
rforma
nce of the
backp
rop
agat
ion n
eural n
e
t
work al
go
rith
m whi
c
h
u
s
in
g
the data of
network wei
ghts with ea
ch
admini
s
tratio
n of learnin
g
rate values
are as sho
w
n in
Table 3.
Figure 4. Testing Interface
from Perfo
r
m
ance
of Backprop
agatio
n Neu
r
al Netwo
r
k Algo
rithm
Table 3. Te
sting Performan
c
e Results of Neu
r
al
Netwo
r
k Algo
rithm
with a Variati
on of Learnin
g
Rate (
α
) Valu
e on Trai
ning
No
Learning Rat
e
(
α
)
Accur
a
cy
Rate
1 0.1
83.33
%
2 0.2
86.67
%
3 0.3
86.67
%
4 0.4
83.33
%
5 0.5
83.33
%
6 0.6
86.67
%
7 0.7
83.33
%
8 0.8
80.00
%
9 0.9
83.33
%
From T
able
2
sho
w
s that t
he be
st a
c
curacy is
obtain
ed on th
e results of trai
nin
g
with a
learni
ng
rate
value
of 0.
2, 0.3, an
d
0.6 in
th
e a
m
ount of
86.
67%.
Whe
r
ea
s the
meth
o
d
of
Suppo
rt Vect
or M
a
chine
s
(SVM) was
used for the
sa
me case
with
the ave
r
ag
e
accuracy
rate
is
85% [7]. This accu
ra
cy level can b
e
improv
ed b
y
incre
a
sin
g
the numbe
r of training
data
becau
se the
performan
ce of backp
ro
pagatio
n neu
ral net
work
algorith
m
is
affected by the
amount of dat
a variation th
at has be
en trained.
4. Conclu
sion
From th
e research th
at h
a
s b
een
don
e,
it can
be
con
c
lu
ded th
at the u
s
e o
f
imag
e
pro
c
e
ssi
ng a
nd ba
ckpropa
gation ne
ural
netwo
rk
te
ch
nique
to cla
s
sify
human h
ead who
a
r
e not
wea
r
ing
a
he
lmet an
d
we
aring
a
helm
e
t provide
s
t
he
g
r
e
a
test a
c
cura
cy rate
of
86.67%.
T
h
is
accuracy lev
e
l is a
c
hieve
d
with a lea
r
ning
rate v
a
lue of 0.2,
0.3 and 0.6
durin
g network
training.
In g
r
anting th
e variation
s
of
learni
n
g
ra
te value during trainin
g
not affect the
perfo
rman
ce
of this neural netwo
rk al
go
rithm.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Cla
ssifi
cation
of Motorcycli
s
ts not Wea
r
Helm
et on Di
gital Im
age With… (Sutikn
o
)
1133
Referen
ces
[1]
W
o
rld He
alth
Organiz
a
tion.
Globa
l Status Rep
o
rt on Ro
a
d
Safet
y
2
0
1
5
. W
o
rld Hea
l
th Organiz
a
tio
n
(W
HO). 2016.
[2]
Statistics Indon
esia. La
nd T
r
ansportati
on
Sta
t
istics 2014. Statistics Indon
e
s
ia. 201
5.
[3] Yudh
o
y
o
no
S
B
.
Regu
lati
on
of the re
pu
blic
of in
don
e
s
ia N
o
2/2
0
0
9
ab
out traffi
c and
roa
d
transportati
on. Jakarta.
200
9.
[4] Coll
ins
on
PA.
T
he ap
plic
ati
on of ca
mera
base
d
traffic mo
nitori
ng s
ystems.
CCT
V and
Roa
d
Surveil
l
a
n
ce, IEE Seminar o
n
.
Londo
n. 200
9: 81-86.
[5]
Parky
n
s DJ,
Bozzo M.
C
C
T
V Ca
mera
S
hari
ng for
Imp
r
oved
T
r
affic Monitori
ng.
Ro
ad
T
r
anspor
t
Information
a
nd C
ontrol-
R
T
IC 2008
an
d IT
S United King
dom M
e
mbers'
Co
nf
erenc
e, IET
.
Manch
e
ster. 2008: 1-6.
[6] Shi
H.
Desi
gn
of a Distribut
ed T
r
affic Mon
i
torin
g
System and Al
gorith
m
Bas
ed o
n
W
eb-Ca
mer
a
.
Information Pr
ocessi
ng (ISIP), 2008 Intern
ati
ona
l S
y
m
posi
u
ms on. Mosco
w
.
2
008: 7
58-7
61.
[7]
Chivert
on J. H
e
lmet pr
esenc
e class
i
ficatio
n
w
i
t
h
motorc
yc
le d
e
tectio
n a
nd tracki
ng
. IE
T Intellig
en
t
T
r
ansport Systems
. 2
012; 6(
3
)
: 259-26
9.
[8]
Azizi A, A
b
b
a
s
pour-Gil
an
de
h Y, N
oos
h
y
a
r
M, Afka
ri-Sa
y
a
h
A. I
dentif
yi
ng
Potato
V
a
rieti
e
s Us
in
g
Machi
ne Visi
o
n
and Artifici
al
Neur
al Net
w
o
r
ks.
Internation
a
l Jour
nal of F
ood Pro
perti
es
.
2016; 19(
3):
618-
635.
[9]
Cheng GJ, Ma
W, Wei X
S
,
Rong
CL, Nan JX. Resear
ch
o
f
ro
ck tex
t
u
r
e i
den
ti
fi
ca
ti
on
ba
sed
on
i
m
a
ge
process
i
ng
an
d ne
ural
net
w
o
rk.
Xi'
an S
h
iy
ou D
a
xu
e Xu
e
bao (Z
ir
an K
e
xue Ba
n)/Jour
nal
of Xi'
a
n
Shiyo
u
Univ
ers
i
ty, Natural Sci
ences Ed
itio
n
. 201
3; 28(5): 10
5-11
0.
[10]
Rajini GK, Reddy
GR.
P
e
rfo
r
ma
nce
eva
l
ua
tion
of ne
ural
netw
o
rks for s
hap
e i
d
e
n
tifica
tion i
n
imag
e
process
i
ng.
Int
e
rnati
ona
l Co
n
f
erence
on Si
gna
l
Acqu
isitio
n an
d Proces
sing (ICSAP
)
.
Bang
alor
e.
201
0: 225-
258.
[11]
Che
n
H,
Xi
ong
L, Hu
X, W
a
n
g
Q, W
u
M. I
dentificati
on m
e
thod for mo
ld
y pea
nut ker
nel
s base
d
o
n
neur
al n
e
t
w
ork
and
imag
e pr
ocessi
ng.
N
ong
ye
Go
ng
che
n
g
Xu
eb
ao
/Tra
n
s
a
c
ti
on
s o
f
th
e
Ch
i
nese
Society of Agri
cultura
l
Engi
ne
erin
g. 200
7
; 23
(4): 158-1
61.
[12]
W
ang Z
,
Cong P, Z
hou J, Z
hu Z
.
Method for ide
n
tifi
catio
n
of
extern
al q
ual
ity of
w
h
e
a
t grai
n base
d
o
n
imag
e pr
ocess
i
ng
an
d artific
i
al n
eur
al
net
w
o
rk.
Non
g
ye
Gongch
e
n
g
X
ueb
ao/T
r
ans
ac
tions of th
e
Chin
ese Soc
i
et
y of Agricultura
l
Engi
neer
in
g
. 200
7; 23(1): 15
8-16
1.
[13]
Yasir R, R
a
h
m
an MA, Ahm
ed N.
D
e
r
m
at
olo
g
ica
l
dis
eas
e detecti
on
us
ing
imag
e pr
o
c
essin
g
an
d
artificial
n
eura
l
netw
o
rk.
8th
Internati
o
n
a
l
Confer
ence
o
n
El
ectrical
a
nd Comp
uter Engi
neer
in
g
:
Advanc
ing T
e
chno
log
y
for a
Better
T
o
morro
w
,
ICECE. P
an Pacific So
n
a
rga o
n
Dhak
a
.
2014: 687-
690.
[14]
Ghali
b
MR, B
h
atnag
ar S, Ja
y
apo
oran
i S, P
a
nde
U. Artifici
a
l
ne
ura
l
n
e
t
w
or
k bas
ed
detect
i
on
of re
nal
tumors usin
g C
T
scan image
process
i
ng.
Int
e
rnati
ona
l Jour
nal of En
gin
eer
ing a
nd T
e
ch
n
o
lo
gy
. 201
4;
6(1): 28-3
5
.
[15]
Hartati S,
Har
j
oko
A, Su
par
di T
W
.
T
he
Digita
l
M
i
cros
cope
a
nd Its
Image
Proc
e
ssing
Util
it
y
.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol.
20
11; 9(3): 565-
574.
[16]
Perman
a
T
P
, Abdur
ohma
n
M, Khairud
i
n
M, Lu
tfi M.
Automated
Na
vigati
on S
y
ste
m
base
d
o
n
W
eapo
n-T
a
rget Assignment.
T
E
LKOMNIKA
T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol
.
201
1; 9(3): 445
-452.
[17]
Aldas
ouq
i I, HassanM. Smar
t Human F
a
ce
Detectio
n S
y
s
t
em.
Internatio
nal Jo
urn
a
l of
Co
mp
uters
.
201
1; 5(2): 201
0-20
17.
[18] Dahab
DA,
Ghoniem
y
SSA,
Selim
GM. Automated Brain
T
u
mor
Detection
and Identification Usi
ng
Image Proces
sing a
nd Prob
abil
i
stic Ne
ural
Net
w
o
r
k T
e
chni
ques.
Inter
natio
nal Jo
urn
a
l of Ima
g
e
Processi
ng an
d Visua
l
Co
mmunic
a
tion
. 2
012
; 1(2): 1-8.
Evaluation Warning : The document was created with Spire.PDF for Python.