Indonesian J
ournal of Ele
c
trical Engin
eering and
Computer Sci
e
nce
Vol. 1, No. 2,
February 20
1
6
, pp. 399 ~
405
DOI: 10.115
9
1
/ijeecs.v1.i2.pp39
9-4
0
5
399
Re
cei
v
ed O
c
t
ober 1
8
, 201
5; Revi
se
d Ja
nuary 14, 20
1
6
; Acce
pted Janua
ry 2
8
, 20
16
IR and Multi Scale Retinex image Enhancement for
Concealed Weapon Detection
Nash
w
a
n J
a
s
im Hussein
*
1
, Fei Hu
2
, Hao Hu
3
, Abd
a
lraza
k
Tare
q Rahem
4
1,2,
3
School of El
ectronic Inform
ation a
nd C
o
m
m
unic
a
tion,
Natio
nal Ke
y
L
abor
ator
y
of Scienc
e an
d T
e
chno
log
y
,H
uaz
h
ong U
n
ivers
i
t
y
of Science a
n
d
T
e
chnol
og
y,
W
uhan 4
3
0
074
, China
4
Departme
n
t of Electrical, Ele
c
tronics an
d Systems En
gin
e
e
rin
g
,
F
a
cult
y
of Engi
neer
ing a
nd Bu
ilt Enviro
nment
, Nationa
l Univ
ersit
y
of Mal
a
ysia UKM
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
1
alsali
hnas
h
w
a
n
@
y
a
h
oo.c
o
m,
2
hufei@m
ail.h
u
st.edu.cn,
3
huha
o.hust@
gmail.c
o
m,
4
abdtareq
@
y
ah
oo
.com
A
b
st
r
a
ct
A Conce
a
l
ed
W
eapo
n Detec
t
ion (CW
D
) ha
d bee
n deve
l
o
ped by a lar
g
e
number of res
earch
ers
and tech
no
log
i
es. As a result of t
he w
eakne
ss of the infrared i
m
a
ges i
n
u
n
iq
ue alt
ogeth
e
r grap
hic ite
m
s,
infrare
d
an
d M
M
W
ima
ges b
e
co
me i
nacc
u
r
a
te an
d ins
u
fficient to o
b
vio
u
s
ly detecta
nd
dea
l w
i
thw
eap
onry
obj
ectsin an i
n
visib
l
e settin
g
. T
h
is article
uses
Multi Scale Reti
nex
and contrast
stretching i
m
ag
e
process
i
ng
e
n
hanc
e
m
ent
te
chni
ques
to
i
m
pr
ove
the
r
e
cogn
ition
of w
eap
ons c
onc
e
a
le
d b
e
l
o
w
attire.
Specific
ally, th
e focus
of the
study is
on
d
e
tect
ing
w
eap
ons a
n
d
a
mmos by
en
hanc
i
ng th
e IR p
i
cture
s
base
d
on
imag
e process
i
n
g
techn
i
qu
es. Evalu
a
tion tec
h
n
i
ques w
e
re e
m
pirica
lly pr
ove
d
to be ab
le to s
how
the enh
anc
e
m
ent perce
ntag
e
progress.
Ke
y
w
ords
: Co
ncea
led W
e
ap
on Detecti
on, IR, and Multi Sc
ale R
e
tinex
Copy
right
©
2016 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
No
wad
a
ys, t
h
reat
s to
h
u
man
safety
are
u
n
c
ea
si
ng to inte
nsifyaround
th
e world.
Therefore, se
curi
ng as
an
efficient way to
en
su
re h
u
m
an
safety is be
comin
g
o
ne of th
e mo
st
seri
ou
s
con
c
ern
s
fo
r auth
o
rities.
Likewise,re
c
e
n
tly, wea
pon
dete
c
tion
ha
s be
come
on
e of
the
most sen
s
itive issue
s
aro
und the worl
d. It has al
so
turned to be
an urg
ent issue that mu
st be
attended to
i
n
term
s of
safety and
se
curity p
u
rp
oses [1]. Rega
rding thi
s
, ma
ny different
ways
and te
chniq
u
e
s
su
ch a
s
hand
gun
s, bl
ade
s, explo
s
ives, ch
emica
l
s an
d as
well as
wea
p
o
n
’s
ammo
shave
been
develo
p
ed for the
pu
rpo
s
e
of dete
c
ting
different
types
of wea
pon
s. Over th
e
years,
several ima
ge fu
si
on m
e
thod
s
have b
een
p
r
opo
sed
to m
eet the
ne
ce
ssitie
s
of alte
red
appli
c
ation
s
.
One
of the
s
e
appli
c
atio
ns that aim
to
detect
all m
u
lti types of
weapo
ns are I
R
sen
s
o
r
s
whi
c
h basi
c
ally u
s
e the temp
e
r
ature sp
re
ad
ing to be spo
tted dire
ctly on the targ
et to
prod
uce an
IR image [2]
.
In fact, IR image
s
are
regul
arly u
s
ed for different purpo
se
s o
f
nightvisio
n
a
pplication
s
, perh
a
p
s
ob
serving
sabl
e
and mova
b
l
e target
s (i.
e
., people
a
n
d
vehicle
s
). Th
e infrared
rad
i
ationemitted
from a
unfix
obje
c
t is en
g
aged by
cloth
i
ng and th
en i
t
is
re-emitted
.
For thisparti
cul
a
r re
ason, the IR image
can be u
s
ed f
o
r presenting
the image of
the
hidde
n targ
et. Howeve
r, d
ue to the dra
w
ba
ck of t
he
IR image
s in
disting
u
ishing
all visual ite
m
s,
IR image
s a
r
e still in
ccurat
e and i
n
suf
fi
cient tocle
a
rly
detecta
nd all
o
cate
or ta
rg
et the locatio
n
of
an invisibl
e
wea
pon. Fo
r
this main
rea
s
on, IR
dete
c
ting device
s
are n
o
t eno
u
gh to dete
c
t the
hidde
n obje
c
t
s
in accu
rate
way and effici
ent detectio
n
.
The use of Conceale
d
We
apon Detecti
on (CWD
) is
basi
c
ally dep
ende
nt on the abilities
of the
hardware
eq
uipme
n
t that
sho
u
ld be
abl
e to
provide
o
r
m
eet somem
a
j
o
r
req
u
ire
m
e
n
ts of
CWD in
cludi
ng being
cap
able of penet
rating an
d de
tecting obje
c
t
s
hidde
n und
er thick cloth
s
,
being a
b
le to
be use
d
fro
m
a long ra
n
ge, and prov
iding the an
al
ysis re
sult im
mediately or
a
t
least si
multa
neou
sly. Yet, hard
w
are
equipm
ents
are u
nable t
o
cove
r and
meetall the
s
e
requi
rem
ents togethe
r at t
he same tim
e
. For i
n
sta
n
c
e,
the millimeter wave sen
s
o
r
(MM
W)
i
s
claime
d to be decently capabl
e of distingui
shi
ng wea
pon
s un
der thick cl
o
t
hs but it hasits
highd
own
s
id
e
in terms of itslon
g ran
ge d
e
tecti
on a
b
ility [3].
In addition, althoug
h infrared sen
s
or
(IR) ha
s be
e
n
able to overcome this
probl
em
or shortcomin
g of MMW,
it sti
ll has drawbacks
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 399 – 405
400
inclu
d
ing it
s
ability topen
etrate thi
c
k
cloth.
Comp
aratively, sof
t
ware
devel
o
p
ment h
a
s b
een
found
and
re
ported
to b
e
much e
a
si
er and
bette
r to meetall
the
req
u
irement
s of
co
nceal
ed
wea
pon
dete
c
tion a
s
co
mpared toth
e hardware
equipm
ent, a
nd it ha
s sh
owe
d
the
sa
me
improvem
ent
s ofthe
ha
rd
ware e
quipm
ent in te
rms
of savin
g
costs, time con
s
uming
and
re
sults
analysi
s
.Man
y IR image enhan
cem
ent techni
que
s h
a
ve been ap
plied to enha
nce IR captu
r
ed
image
s li
ke
d
ouble
-
de
nsity
dual
-tre
e
co
mplex wavele
t tran
sform
(DDDT
C
WT) [
4
], homom
orphic
filter [5], SIFT, Histo
g
ra
m
thre
shol
ding
and m
a
tc
h
e
d
filtering [6].
Lea
rning
Ve
ctor
Qua
n
tiza
tio
n
(LVQ
) net
wo
rk p
r
op
osed
b
y
[5] inclu
ded
de
signi
ng a
hard
w
a
r
e
hig
h
freq
uen
cy
mech
ani
sm,
but
the disadvant
age of this
m
e
thod i
s
its hi
gh cost
of bui
lding an
d de
signing the
mo
del. The
r
efore,
based o
n
all t
hese p
r
eviou
s
ly highli
ghte
d
issue
s
,this work rep
o
rte
d
in
the cu
rre
n
t
pape
rpropo
se
s
a better
sol
u
tion to
solve t
he issu
esand
dra
w
b
a
cks o
f
the previo
u
s
meth
od
s by
introd
ucin
g t
h
e
appli
c
ation
of
a m
o
re effe
ctive and
bette
r d
e
tectio
n m
e
thod to
enh
a
n
cin
g
the
cap
t
ured
imag
e,
as
explained b
e
l
o
w.
In this resea
r
ch,
we
pre
s
ent a ne
w m
e
thod to
enh
ance the
ca
ptured
IR im
age
s by
applying M
u
lti scale
Retine
xon a V col
o
r cha
nnel. Fo
r this, the the
V colo
r chan
nel was
divided
into non
-ove
rlappi
ng bl
ocks
by the si
ze of
6
4
x64
,
and blo
c
k wa
s enh
an
ced
se
parately.
More
over,
we
mea
s
ured th
e features
of each bl
o
c
k
a
nd
ap
plied a method calle
d
“exp
osure
”
to
disting
u
ish b
e
twee
n the d
a
rk
and the
bright regi
o
n
s
in o
r
de
r to be able to d
e
tect the hid
den
obje
c
t more
clea
rly. We
a
l
so ma
de the
dark regio
n
darke
r an
d th
e brig
ht re
gio
n
brig
hter to
be
able to highli
ght the hidde
n obje
c
t.
2. Related Work
Many previo
us research
e
r
s h
a
ve pro
posed an
d develop
ed i
m
age
-rel
a
ted
tools or
techni
que
s fo
r enh
an
cing
con
c
e
a
led
weapo
n dete
c
ti
on. An autom
atic re
gist
rati
on algo
rithm
for
IR an
d MM
W image
s
wa
s
been
present
ed by [1]. Thi
s
meth
od i
s
b
a
se
d o
n
the
work
rigi
d bo
dy
transformation. Otherwi
se,
it will
not be
able to make the whol
e body visible. While other authors
havede
alt with X-Ray ima
ges d
a
taba
se
to detect hidden obje
c
ts [
7
], this method only focu
sed
on dete
c
ting
a few type
s
of ammo
s u
s
ing si
mp
le i
m
age e
nha
nce
m
ent techniq
ue, and
ign
o
red
detectio
n
of a
ll other types
of object
s
. In [4
], the rese
arche
r
s int
r
o
duced two d
e
c
isi
on meth
o
d
s
whi
c
h
sig
n
ificantly imp
r
o
v
e the ima
g
e
fusi
on
perf
o
rma
n
ce for co
nceale
d
wea
pon
dete
c
tion
appli
c
ation. F
u
rthe
rmo
r
e,
standoff di
stan
ce i
s
sue
du
ri
ng the
ru
sh
h
our i
n
ai
rpo
r
t
s
wa
s solved
b
y
[5] by applying hom
omo
r
phic filter
on
blocks an
d
blendi
ng for i
m
age fu
sion.
Artificial Ne
ural
Network ANN method
ba
sed on
the P
C
A wa
s introd
uce
d
by [8]. A three
dime
nsio
nal n
ear field
imaging
algo
rithm also
use
d
to com
pare
near fiel
d si
mulation
wa
sprop
o
sed by
[9]. An automatic
detectio
n
an
d
re
cog
n
ition system
of con
c
eal
ed wea
p
ons
u
s
ing se
nso
r
te
chn
o
lo
gies
and
ima
ge
processi
ng was introduced by [
10]. Milli
meter-wave i
m
aging
Radi
ometer Equipment (MIRAE
) by
applying a di
electri
c
a
s
ph
eric le
ns a
n
d
a metal
mirror as a reflector, 30 cha
n
n
e
ls and an F
P
A
receiver b
a
se
on the conve
r
sio
n
type wa
s pro
p
o
s
ed b
y
[11].
3. Methodol
og
y
Con
c
e
a
led weapo
n detecti
on might not detect a cert
ain wea
pon a
nd it will only be able
to make
su
ch
weap
on ap
p
ear in the
ca
ptured im
age
. Thus, mea
s
uring th
e hist
ogra
m
of eve
r
y
captu
r
ed ima
ge doe
s not occupy
the whole dynami
c
range. The
c
apture
d
imag
es u
s
ually su
ffer
from an ove
r
-saturation p
r
oblem, lo
w contra
s
t, non-uniform illu
mination an
d u
neven lighte
n
i
ng.
All these problem
s affectthe ability to detec
t a
hidden object. Theref
ore, measuring the
histog
ram of
the captu
r
e
d
image that
sho
w
s
th
e unde
rlying int
ensity expo
si
tion may occupy
more
of the
l
o
we
r p
a
rt
or
the up
per p
a
r
t of th
e
total
ra
nge
of the
histo
g
ram of
the
captu
r
e
d
image.
Whe
n
the g
r
ay le
vels of th
e
capture
d
im
ag
e contain
m
o
st of th
e lo
wer pa
rt of t
h
e
histog
ram a
r
ea, then, the region
appe
ars
dark whil
e it appea
rs
very
bright when its cray l
e
vel
occupi
esm
o
re of the upp
er of the hist
ogra
m
. So
me
parts of the image might
also
suffer from
being
overex
posed regio
n
s
whe
r
ea
s ot
her
part
s
of
t
he imag
e
can
be un
de
rexp
ose
d
at the
same
time. Hen
c
e,
the imag
e a
r
ea th
at it su
ffers
from a
n
overexp
o
se
d re
gion
or
a
n
und
erexpo
sed
regio
n
m
a
y carry l
e
ss i
n
formation tha
n
t
he ima
ge
wit
h
a
well
-expo
s
ed
re
gion.
T
o
overco
me t
h
is
probl
em, we f
u
sed the exposure
of the
captured im
age so that
the dynamic
range of it.In would
be en
han
ce
d. Figure 1 ill
u
s
trate
s
the
proce
s
sing
ste
p
s of th
e p
r
o
posed m
e
tho
d
followed in
the
cur
r
e
n
t
st
udy
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
IR and Multi Scale Retine
x im
age Enha
ncem
ent for
Con
c
e
a
led…
(Na
s
h
w
a
n
Ja
sim
Hussein
)
401
3.1. The Proposed Me
th
od
G
e
n
e
r
a
lly, to
p
r
oc
es
s th
e ca
p
t
u
r
ed
vis
u
al ima
g
e
,
it
w
a
s
c
o
n
v
er
te
d in
to
H
S
V co
lor
s
p
ac
e
by se
paratin
gthe
chromat
i
c info
rmatio
n and
the
a
c
hromati
c
inf
o
rmatio
n, an
d leaving
th
e IR
comp
uted im
age a
s
it i
s
. T
he p
r
opo
se
d
method
of this stu
d
y was
applie
d follo
wing the follo
wi
ng
eight step
s:
1.
Conve
r
ting th
e input visual
image in to HSV color
spa
c
e.
2.
Enhan
cingth
e
IR imag
e a
nd the visual
image
usi
n
g the MS
R
method th
at can
wa
sho
u
t the degrade
d ima
ge and the n
o
n
-unifo
rm ligh
t
ing.
3.
Dividingthe o
u
tput enha
nced image
into
64x64 non
-o
verlappi
ng bl
ocks.
4.
Usi
ng th
e V
space
colo
r f
r
o
m
the
HSV
a
nd fu
singit
to
gether
with th
e en
han
ce
d i
nput
IR image.
5.
Extracting the
feature
s
of
each bl
ock of the fuse
d ima
ge.
6.
Determinin
g
or
disting
u
ish
i
ng b
e
twe
en
the da
rk a
n
d
bri
ght regio
n
to en
han
ce
the
over and u
n
d
e
r-expo
sed p
r
oble
m
for ea
ch blo
c
k usi
n
g the exposure method.
7.
Extracting the
histogram of the di
vided i
m
age u
s
ing
contra
st stret
c
hing.
8.
Re-co
m
binin
g
the thre
e chann
els, HS
V,conv
ertin
g
it back to th
e
origin
al form
and
getting or obt
aining the e
n
han
ced o
u
tpu
t
image.
These step
s
are explai
ned
in more detai
ls as follo
ws:
Figure 1. The
propo
se
d me
thod pro
c
e
s
si
ng
3.2. Multi Scale Retin
e
x
Multi Scale Retinex is a method used to
enhan
ce the image
s whi
c
h
areaffecte
d
b
y
a low
contrast, une
ven contrast
and illumin
a
tion. This
met
hod ba
si
cally works o
n
two major fa
cto
r
s
whi
c
h are: illumination a
n
d
reflecta
nce. Usin
g
the Multi scal
e
Reti
nex method
as propo
se
d and
pre
s
ente
d
by [12-17], the image is
com
posed by
two
major pa
rts
whi
c
h are light and refle
c
tance
of the object
as sho
w
n in
Equation (1):
/
(1)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 399 – 405
402
Whe
r
e, L: the value of incident light; R:
the value
of
obje
c
t’s refle
c
tion; E: the
value of reflecte
d
light.
MSR is a
co
mbination of
the weig
hted
sum of t
he o
u
tput of different sizes
of scale
s
of
Single Scale
Retinex (SSR) [18-2
0
]. MSR wa
s used
i
n
this study to enhan
ce th
e input image
by
solving the lo
w co
ntra
st
an
d uneven i
s
sues of it.
MSR ca
n be i
llustrate
d in Equation (2):
RMSRi
=
∑
=
∑
[log (Ii(x
,
y
))
−
log (
I
i(
x, y)
∗
Fn(x
, y
))]
(2)
Ri(x, y): is
the Retinex out
put; i:
∈
R,G,
B col
o
r chan
nels; Ii
(x, y): i
s
the
in
put im
age
whi
c
h
be
en
distrib
u
ted a
m
ongthe th
re
e cha
nnel
s RGB.F(x, y): th
e surro
und fu
nction.
*: mathematical convolution
operation
bet
wee
n
Ii(x,
y)
and F(x, y); w
here
,
can be
explained in
Equation (3).
,
ex
p
/
(
3
)
,
: The surro
u
n
d
function.
: is the Gaussian di
strib
u
tion function
whic
h was fixe
d by the authors to 15, 80
and
250.
: is the norma
lized fa
ctor
∬
,
1
.
Figure 1
sho
w
s th
e dia
g
ram of MSR
a
l
gorithm
whi
c
h used the
G
aussia
n
fun
c
tion with
t
he t
h
ree
scal
e
siz
e
s.
Figure 1. Multi-Scal
e Retin
e
x block diag
ram
3.3. Div
i
ding
into Blocks
and Mea
s
uri
ng Feature E
x
tra
c
tion
After MSR a
pplication to the input im
a
ges
(ove
r-ex
posed an
d u
nder-exp
osed
image
s)
wa
s a
c
compl
i
she
d
, this
stage focused
on regul
atin
g the b
e
st
exposed im
a
ges. T
h
is was
con
d
u
c
ted by isolating the
image into non-over
l
appi
ng 64x64 blo
ck
size and
cla
ssifying e
a
ch
block se
pa
rat
e
ly into overexpose
d
, unde
rexpo
s
ed a
n
d
well-exp
o
se
d blocks u
s
in
g the exposu
r
e
[21].
3.4. Featur
e Extrac
tion
This
wa
s a
c
hieved by m
easurin
g the
featur
e
s
of
each blo
c
k separately usi
ng thre
e
intensity feat
ure
s
a
nd
co
mputing th
e
m
for e
a
ch
b
l
ock for the
whol
e ima
ge.
The fe
ature
s
a
r
e
minimum inte
nsity, maximum inten
s
ity and ave
r
ag
e
i
n
tensity. The
feature
extra
c
tioni
s illu
stra
ted
in the followin
g
Equation
s
(4,5,6):
min
∑
(
4
)
max
∑
(
5
)
∑
∑
(
6
)
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
IR and Multi Scale Retine
x im
age Enha
ncem
ent for
Con
c
e
a
led…
(Na
s
h
w
a
n
Ja
sim
Hussein
)
403
Whe
r
e, q: the
numbe
r of the distin
ct gra
y
level
of each block; p(q
)
:
image histo
g
r
am; l: intensi
t
y
level
The average
intensity is the averag
e pi
xel value whi
c
h is
able to dete
r
mine the
brightn
e
ss o
r
darkne
s
s of the blocks while
the mini
mum and m
a
ximum intensities are able
to
define the mi
nimum an
d maximum inten
s
ities valu
e of each bl
ock.
After mea
s
uring the exp
o
s
ure, minimu
m in
tensity, maximum int
ensity an
d a
v
erage
intensity of
e
a
ch
blo
c
k, th
en e
a
ch
block
woul
d
have
its own
exp
o
s
ed
a
r
ea, eith
er over, und
e
r
o
r
well-expo
sed.
For thi
s
, we u
s
e
d
or applie
d
co
ntrast
stret
c
hing to the
overexpo
se
d or
unde
rexpo
s
e
d
blocks onl
y based o
n
the histog
ra
m of each b
l
ock whil
e ignorin
g the well-
exposed blo
c
ks. Co
ntra
st
stret
c
hing
wa
s u
s
e
d
to
ch
ange
the
bri
g
htness l
e
vel f
o
r th
e
sele
ct
e
d
blocks only b
a
se
d on this
Equation (7):
255
(
7
)
Whe
r
e ,
and
values are ta
ken from equ
ation
s
(4,5).Th
e ou
tput of each
enha
nced blo
ck
wa
s com
b
i
ned togeth
e
r
to build t
he o
u
tput image, and the H, V and S cha
n
n
e
ls
were co
mbin
ed togethe
r to get the outp
u
t enhan
ce
d image.
4. Results a
nd Discu
ssi
ons
This
sec
t
ion disc
us
ses
t
he major
results
of
ou
r
e
x
pe
r
i
me
n
t
th
a
t
focu
s
e
d
on
pr
opo
s
i
ng
th
e
adaptive
enh
ancement
pa
rtition blo
c
ks based on M
S
R
a
nd co
ntrast
stret
c
hin
g
.
As p
r
eviou
s
ly
s
t
a
t
e
d
,
th
e pro
p
o
s
e
d me
tho
d
ba
s
i
c
a
lly
w
o
rks
on
two
types
of ima
ge: infra
r
e
d
a
nd visual im
a
ges.
It is able to
tran
sform
the
input visual
image i
n
to th
e hue
-saturation-valu
e (HSV) col
o
r
sp
ace
and enh
an
ce
the V channel whi
c
h on
ly contains
t
he brig
htne
ss informatio
n
of the images.
Based
on th
e re
sults,
ap
plying the M
u
lti Scal
e
Re
tinex enha
ncement meth
o
d
co
uld a
c
hi
eve
both: enhan
ci
ng the input image
s and restori
ngthe
d
egra
ded ima
g
e
by non-unif
o
rm illumin
a
tion.
Dividing the i
m
age into 6
4
x
64 non
-overl
ap blo
c
ks e
n
abled u
s
to d
e
termin
e or d
i
stingui
sh a
m
ong
the over-exp
ose
d
, und
er-exposed
a
nd
well-expo
sed
blocks by a
p
plying the
exposure, mi
ni
mum
intensity, ma
ximum intensity and average inten
s
ity that could
d
e
termin
e the
dark an
d bright
regio
n
. In this stu
d
y, imag
e fusio
n
was
applie
d by co
mbining th
e three
ch
ann
el
s of HSV
col
o
r
spa
c
e
and
co
nverting it
ba
ck into the
ori
g
inal
fo
rm to
have the
outp
u
t enha
nced i
m
age. Fi
gure
3
sho
w
s the ori
g
inal imag
e b
o
rrowed from
[5], HSV
ima
ge and the hi
stogram of ea
ch on
e of them.
Figure 3. Orig
inal image in
RGB and
HS
V color
spa
c
e
and its histo
g
ram
In com
pari
n
g between
the imag
e
befor
e
en
ha
ncem
ent a
n
d
after
enh
ancement
byapplying M
S
R and
co
ntrast st
retching
to the out
pu
t image (Fi
g
u
r
e 4
)
, it is evident that the
entropy
mea
s
ureme
n
t for the o
r
iginal
visual i
m
ag
e wa
s5.3
535
while
the
e
n
tropy fo
r th
e
enha
nced im
age imp
r
ove
d
to be 5.1
068
. Our exp
e
ri
ment
provide
s
evide
n
ce of
the ability of
the
prop
osed met
hod to enh
an
ce or im
prove
the IR origin
al image from
0.13229 to 0.
0607
9.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 399 – 405
404
Table 1. Prop
ose
d
method
results
Me
t
h
od En
t
r
op
y
Origi
n
al im
age
5.3535
IR metho
d
5.2265
Propose
d
m
e
th
od
5.1068
Figure 2. IR image befo
r
e
enha
ncm
ent and a
fter en
h
ancm
ent usi
n
g MSR and
Contra
st
Streaching
Figure 3. Pro
posed metho
d
pro
c
e
ss
sta
ges
5. Conclusio
n
In this
pap
er,
we
presente
d
a
pro
p
o
s
ed
algo
rithm fo
r image
en
han
ceme
nt of
co
nce
a
led
wea
pon dete
c
tion.
Bein
g motivated
to solve se
riou
s issue
s
in
clu
d
ing n
o
n
-
unif
o
rm, lo
w
cont
rast
over-saturation problem
s
f
r
om
whi
c
h infrared and visual im
age
s
still suffer, the Multi Scal
e
Retinex alg
o
rithm wa
s pro
posed an
d a
pplied in th
i
s
study. This
was
carrie
d ou
t by dividing the
image into bl
ocks to dete
r
mine the dark and the
b
r
i
ght regio
n
a
nd usin
g 64x
64 non
-overl
ap
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
IR and Multi Scale Retine
x im
age Enha
ncem
ent for
Con
c
e
a
led…
(Na
s
h
w
a
n
Ja
sim
Hussein
)
405
blocks to e
n
h
ance ea
ch
bl
ock individ
ual
ly. The
re
sult
s sho
w
ed
tha
t
applying co
ntrast stretchi
ng
wa
s able to solve the overexpos
ed and
unde
r-expo
se
dimage i
s
sue
s
. In addition, convertin
g
the
visual imag
e into the HSV colo
r sp
ace wa
s prov
e
d
tobe abl
e to provide faste
r
pro
c
e
ssi
ng time
by enha
nci
n
g
theV space
only co
mpa
r
e
s
to the
u
s
e
of the RGB color
sp
ace. T
he expe
rime
n
t
al
results in
dica
ted that ou
rpropo
sed m
e
th
od was
able
to gene
rate b
e
tter re
sult
compa
r
ed to t
he
image
s ta
ken
by the I
R
se
nso
r
s,
or MM
W
sen
s
o
r
s. In con
c
lu
sion,
ou
r expe
rim
ent ine
nha
nci
n
g
the visual im
age an
d the
IR image u
s
ing Multi S
c
ale
Retinex
and co
ntra
st stretchin
g
has
been
su
cce
ssf
ully achieve
d
by applying t
he expo
su
re,
enablin
g u
s
to determi
ne
the dark regi
o
n
and b
r
ight re
gion by ma
ki
ng the bri
ght regio
n
bri
ghte
r
and the d
a
rk re
gion d
a
rker. This
wo
rk
can
be furthe
r de
veloped to a
nother
stag
e
using
dete
c
tion metho
d
s
like ap
plying
artificial ne
u
r
al
netwo
rks techniqu
es.
Referen
ces
[1]
Varshney
PK, et al.
Registrat
i
on a
nd fusio
n
of infrared an
d mi
lli
meter wave i
m
ag
es for conceal
e
d
w
eapon
d
e
tection.
in
Image
Processi
ng
. IC
IP 99. Pr
ocee
din
g
s. 1
999
In
ternatio
nal
C
o
nferenc
e
on
.
IEEE. 1999.
[2]
Z
ou Y, et al. Visibl
e an
d Infrared Image F
u
s
i
on Usi
ng the
Lifting W
a
ve
let
.
T
E
LKOMNIKA Indones
ia
n
Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 201
3; 11(1
1
): 6290-6
2
9
5
.
[3]
Yang B
h
, et al. Desig
n
of Rec
e
iver Us
ed for
Passive Mi
llim
eter W
a
ve Imagin
g
S
y
stem.
TE
L
K
OM
N
I
KA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2014; 1
2
(1): 9
8
-10
5
.
[4]
T
u
zhi X
,
QMJ Wu.
Multisen
sor conc
eal
ed
w
eapo
n d
e
te
ction us
in
g th
e i
m
a
ge fus
i
o
n
ap
pro
a
ch
. in
Imagin
g
for Cri
m
e Preventi
on
and D
e
tectio
n (ICDP
-15). 6th Internati
o
n
a
l Co
nferenc
e on. 2
015.
[5
]
U
p
ad
hy
ay
EM, N
K
R
a
na
.
Exposur
e fusio
n
for conce
a
l
ed
w
eapon
detecti
on. in D
e
vices,
Circuits an
d
System
s (ICDCS)
. 2nd Intern
ation
a
l Co
nfere
n
ce on. 2
014.
[6]
Carrer L,
AG Yarovoy
.
Co
nc
eal
ed
w
e
apo
n detectio
n
usin
g
UW
B
3-D r
a
dar
i
m
ag
in
g a
nd
auto
m
ati
c
target recog
n
iti
on. in Ante
nna
s and Prop
ag
ation (Eu
C
AP).
8th Europe
an C
onfere
n
ce o
n
. 201
4.
[7] Al-Najda
w
i
N.
A conc
ealed amm
o
detection system
for
pa
ssengers luggage scre
ening.
in Multimedia
Co
mp
uting a
n
d
Systems (ICMCS).
Internatio
nal C
onfere
n
ce
on. 201
4.
[8]
O'Reill
y D,
N
Bo
w
r
in
g,
S Ha
rmer.
Sig
nal
pr
ocessi
ng tec
h
n
i
qu
es for c
onc
eal
ed w
e
apo
n
detectio
n
b
y
use
of ne
ural
n
e
tw
orks
. In Electrical & E
l
ectr
onics E
n
gineer
s
in Israel (IEE
EI).
IEEE 27th
Conv
ention
of. 2012.
[9]
Corucci
L, et al
. Near F
i
eld P
h
ysic
al Optics m
ode
lli
ng for Co
ncea
led W
e
ap
on Detecti
on.
Antenn
as an
d
Propagation, IEEE
Transactions on
. 2
012;
60(1
2
): 605
2-6
057.
[10]
Band
yo
p
adh
ya
y P, et al.
Ident
ificatio
ns of concea
l
ed w
e
a
p
o
n
in a Hu
man
Body
. 201
2.
[11]
W
on-G
y
um
K,
et al.
Syste
m
d
e
sig
n
of foc
a
l
p
l
an
e arr
a
y b
a
s
ed Mi
lli
meter-w
ave I
m
a
g
in
g r
a
dio
m
eter for
conce
a
le
d w
eapo
n det
ectio
n
.
In Geoscie
n
ce a
nd R
e
m
o
te Sens
ing S
y
mp
osi
u
m (IGARSS). IEEE
Internatio
na
l. 2011.
[12] Lan
d EH.
An
a
l
ternativ
e tech
niq
ue for th
e c
o
mputati
on
of
the d
e
sig
nator
i
n
the reti
nex th
eory of co
lor
vision.
Proc
ee
din
g
s of the Na
tiona
l Acad
emy
of Scie
nces. 198
6; 83(1
0
): 3078-
308
0.
[13]
Lan
d EH. Rece
nt advanc
es in
retine
x th
e
o
r
y
.
Visio
n
researc
h
. 1986; 2
6
(1): 7-21.
[14]
Lan
d EH. T
he retine
x the
o
r
y
o
f
color visio
n
.
Scientific A
m
er
ic
a
. 1977.
[15] Lan
d
EH.
Rec
ent adv
anc
es i
n
retin
e
x the
o
ry and s
o
me i
m
p
licati
ons for
cortical co
mp
utations: co
l
o
r
vision
an
d th
e
natura
l
i
m
a
ge.
Procee
din
g
s of
the N
a
tion
al A
c
adem
y
of Sci
ences
of
the U
n
ited
St
ates
of America. 19
83; 80(1
6
): 516
3.
[16]
Lan
d EH, J McCan
n
. Lig
h
tnes
s and retin
e
x
theor
y
.
JOSA
, 1971; 61(
1): 1-1
1
.
[17]
Lan
d EH, JJ Mccann. Meth
od an
d s
y
st
e
m
for im
age
repro
ductio
n
b
a
sed o
n
sig
n
i
f
icant visua
l
bou
nd
aries of
origi
n
a
l
subj
ect
.
Google Pate
n
t
s
. 1971.
[1
8
]
Ki
ra
gu
H
,
E Mw
ang
i
.
An
i
m
p
r
oved
en
ha
nce
m
e
n
t of d
egr
a
ded
bi
nary tex
t
docu
m
e
n
t i
m
ages
usi
n
g
mor
p
h
o
lo
gica
l
and s
i
n
g
le
scal
e
retin
e
x
oper
ations.
In Im
ag
e Process
i
n
g
(
I
PR 20
12), IET
Confer
enc
e
on. 201
2.
[19]
Ron
g
Z
,
W
L
Jun. Improv
ed
w
a
v
e
l
e
t transform al
gor
i
t
hm for singl
e imag
e d
e
h
a
zin
g
.
Optik -
Internatio
na
l Journ
a
l for Li
ght
and Electro
n
Optics.
2014; 1
25(1
3
): 306
4-3
066.
[20] Karimimehr
S.
A novel fac
e
recognition system
ins
p
ired by com
p
utational neur
oscience
. I
n
EUROCON, IEEE
. 2013.
[21]
Hanm
and
lu M, et al.
A nove
l
opti
m
a
l
fu
z
z
y
system for color i
m
a
ge e
n
hanc
e
m
ent us
i
ng bacter
i
a
l
foragi
ng.
Instru
mentatio
n an
d Measur
ement, IEEE
T
r
ansactions on. 2
009;
58(8): 28
67-
28
79.
Evaluation Warning : The document was created with Spire.PDF for Python.