Inter
national
J
our
nal
of
Electrical
and
Computer
Engineering
(IJECE)
V
ol.
7,
No.
5,
October
2017,
pp.
2581
–
2595
ISSN:
2088-8708
2581
I
ns
t
it
u
t
e
o
f
A
d
v
a
nce
d
Eng
ine
e
r
i
ng
a
nd
S
cie
nce
w
w
w
.
i
a
e
s
j
o
u
r
n
a
l
.
c
o
m
Intensity
Pr
eser
ving
Cast
Remo
v
al
in
Color
Images
Using
P
article
Swarm
Optimization
Om
Prakash
V
erma
1
and
Nitin
Sharma
2
1
Delhi
T
echnological
Uni
v
ersity
,
Delhi,
India
2
Maharaja
Agrasen
Institute
of
T
echnology
,
Rohini,
Delhi,
India
Article
Inf
o
Article
history:
Recei
v
ed:
Jan
18,
2017
Re
vised:
Jun
12,
2017
Accepted:
Jun
30,
2017
K
eyw
ord:
Gamma
correction
color
cast
P
article
sw
arm
optimization
Intensity
preserv
ation
Knee
transfer
function
ABSTRA
CT
In
this
paper
,
we
present
an
optimal
i
mage
enhancement
technique
for
color
cast
images
by
preserving
their
intensity
.
There
are
methods
which
impro
v
es
the
appearance
of
the
af
fected
images
under
dif
ferent
cast
lik
e
red,
green,
blue
etc
b
ut
up
to
some
e
xtent.
The
proposed
color
cast
method
is
corrected
by
using
transformation
funct
ion
based
on
g
amma
v
alues.
These
optimal
v
alues
of
g
amma
are
obtained
through
particle
sw
arm
optimization
(PSO).
This
technique
preserv
es
the
image
intensity
and
maintains
the
originality
of
color
by
satisfying
the
modified
gray
w
orld
assumptions.
F
or
the
performance
analysis,
the
image
distance
metric
criteria
of
CIELAB
color
space
i
s
used.
The
ef
fecti
v
eness
of
the
proposed
approach
is
illustrated
by
t
esting
the
proposed
method
on
color
cast
images.
It
has
been
found
that
distance
bet
ween
the
reference
image
and
the
corrected
proposed
image
is
ne
gli
gible.
The
ca
lculated
v
alue
of
image
distance
depicts
that
the
enhanced
image
results
of
the
proposed
algorithm
a
re
closer
to
the
reference
images
in
comparison
with
other
e
xisting
methods.
Copyright
c
2017
Institute
of
Advanced
Engineering
and
Science
.
All
rights
r
eserved.
Corresponding
A
uthor:
Nitin
Sharma
MAIT
,
GGSIPU
Rohini
Delhi,India
Email:
sharmaisnitin@gmail.com
1.
INTR
ODUCTION
Color
cast
remo
v
al
is
a
challenging
task
under
dif
ferent
illumi
nation
conditions.
Image
captured
by
digital
camera
are
usually
depend
on
v
arious
properties
of
the
de
vice
and
source
of
the
illumination.
The
major
adjustment
is
requi
red
on
the
color
content
and
intensity
of
the
image.
The
color
cast
is
due
to
the
color
of
ambient
light
which
sho
ws
lo
w
or
high
contrast,
o
v
er
e
xposure
or
under
e
xposure
of
some
re
gions
lead
to
the
dif
ference.
These
major
causes
are
remo
v
ed
with
the
proposed
approach.
The
image
enhancement
may
be
carried
out
by
increasing
the
image
contrast.
The
contrast
increment
can
be
achie
v
ed
by
using
v
arious
algorithms
lik
e
histogram
equalization,
global
histogram
equalization
etc
[1];
[2].
In
these
methods
generally
,
the
image
is
enhanced
b
ut
its
information
content
gets
reduced
significantly
[3].Buchsbaumet
[4]
has
proposed
the
gray
w
orld
assumption
based
method
for
color
constanc
y
in
a
de
graded
image.
It
meets
the
criterion
of
human
visual
system.
T
ang
[5]
ha
v
e
presented
a
method
to
enhance
the
color
image
by
di
viding
it
into
chromaticity
and
intensity
components.
Kw
ok
[16]
a
v
oids
the
color
saturation
by
modifying
the
gray
w
orld
assumption.
F
arid
[7]
proposed
the
gray
image
enhancement
by
using
g
amma
correction.
Monobe
[8]
used
the
knee
transfer
function
based
g
amma
correction.
Finding
an
optimal
v
alue
of
g
amma
is
al
w
ays
a
dif
ficult
task.
Ev
olutionary
algorithms
ha
v
e
been
used
to
perform
image
enhancement
[9];
[19];
[11].
One
of
the
main
dra
wbacks
of
the
pre
viously
used
e
v
olutionary
algorithms
is
the
lack
of
memory
a
v
ailability
which
limits
its
search
and
con
v
er
gence
ability
.
Guan
[12]
ha
v
e
discussed
the
application
of
GA
to
determine
the
g
amma
v
alue.
In
the
proposed
method,
we
ha
v
e
used
PSO
for
optimizing
the
g
amma
v
alue
used
in
image
contrast
enhancement.
In
comparison
to
GA,
PSO
[13]
is
simple
and
has
less
comple
xity
as
it
does
not
require
the
selection,
crosso
v
er
and
mutation
operations
that
are
in
v
olv
ed
in
GA.
PSO
has
fe
wer
parameter
and
f
ast
con
v
er
gence
rate
as
it
does
not
use
the
survi
v
al
of
the
fittest
concept.
The
particle
ha
ving
lo
wer
fitness
can
survi
v
e
during
the
optimization
and
J
ournal
Homepage:
http://iaesjournal.com/online/inde
x.php/IJECE
I
ns
t
it
u
t
e
o
f
A
d
v
a
nce
d
Eng
ine
e
r
i
ng
a
nd
S
cie
nce
w
w
w
.
i
a
e
s
j
o
u
r
n
a
l
.
c
o
m
,
DOI:
10.11591/ijece.v7i5.pp2581-2595
Evaluation Warning : The document was created with Spire.PDF for Python.
2582
ISSN:
2088-8708
potentially
visit
an
y
point
in
the
search
space
[14];
[15];
[16];
Kw
ok
[17];
[18].
In
the
proposed
method,
the
PSO
is
used
for
finding
the
optimal
v
alue
of
g
amma
by
preserving
mean
intensity
v
alues.
The
method
enhances
the
color
images
ef
fecti
v
ely
and
automatically
without
prior
illumination
kno
wledge.
In
this
paper
a
no
v
el
method
is
proposed
which
uses
single
fitness
function.
It
utilizes
the
PSO
and
gi
v
es
optimal
v
alue
under
non-linear
conditions.
The
paper
is
as
or
g
anized
as
follo
w
.
The
proposed
fitness
g
amma
correction
method
based
on
knee
transfer
function
is
introduced
in
Section
2.
Section
3
discusses
the
PSO
algorithm.
Later
in
Section
4,
the
proposed
algorithm
is
de
v
eloped
for
color
cast
remo
v
al.
The
per
formance
measures
and
results
are
discussed
in
section
5
section
6
respecti
v
ely
.
Finally
,
conclusions
dra
wn
from
the
results
obtained
are
mentioned
in
Section
7.
2.
MODIFIED
GAMMA
CORRECTION
B
ASED
ON
KNEE
TRANSFER
FUNCTION
In
the
present
application
we
ha
v
e
considered
Red
(R)
Green
(G)
and
Blue
(B)
color
model
of
color
space.
The
mean
v
alue
R
,
G
and
B
of
red,
green
and
blue
channel
respecti
v
ely
of
a
color
image
of
size
MxN
is
gi
v
en
by
R
=
1
MN
M
X
i=1
N
X
j=1
R(i
;
j)
(1)
G
=
1
MN
M
X
i=1
N
X
j=1
G(i
;
j)
(2)
B
=
1
MN
M
X
i=1
N
X
j=1
B(i
;
j)
(3)
Where
i
and
j
denotes
the
indices
of
pix
el
position.
The
mean
intensity
of
an
image
is
gi
v
en
by
=
(
R
+
G
+
B)
=
3
(4)
R
channel,
G
channel
and
B
channel
are
normalized
in
such
a
w
ay
that
each
channel
has
its
v
alue
lying
in
the
range
[0,
1].
Gamma
correction
is
a
nonlinear
adjustment
met
h
od
used
for
color
correction.
W
e
define
the
modified
g
amma
correction
method
based
on
knee
transfer
function
(say
for
red
channel)is
obtained
by
modifying
the
con
v
entional
knee
curv
e
as
gi
v
en
by
I
=
8
>
>
<
>
>
:
1
256
256
255
P
i=0
255
P
j=0
R(i
;
j)
;
if
R
<
t
1
256
256
255
P
i=0
255
P
j=0
(a
1
R
3
(i
;
j)+a
2
R
2
(i
;
j)+a
3
R(i
;
j)+a
4
)
;
elsewhere
(5)
where
I
is
the
output
intensity
le
v
el
after
the
g
amma
correction
and
t
denotes
the
threshold
le
v
el
for
each
channel
which
is
tak
en
as
0
:
35
in
our
case.
The
normalize
d
v
alues
are
raised
to
the
po
wer
of
as
^
R
=
R
;
^
G
=
G
;
and
^
B
=
B
after
comparing
the
mean
v
alues
for
each
channel
wi
th
the
threshold
v
alue
and
are
a
function
of
.
If
>
1
then
a
v
erage
intensi
ty
v
alue
increases
and
vice
v
ersa.
There
are
methods
lik
e
gray
w
orld
assumption
which
assumes
mean
v
alue
for
correction
f
actor
.
Here,
mean
intensity
of
each
channel
is
decrement
or
increment
for
optimal
v
alue
of
g
amma
and
equal
to
the
aggre
g
ated
mean
intensity
of
the
image.
In
case
the
mean
v
alue
of
channel
is
found
higher
than
the
selected
threshold
v
alue
’
t’,
then
the
approximate
con
v
entional
knee
curv
e
transforms
the
intensity
from
linear
curv
e
to
the
cubic
curv
e
at
the
same
selected
threshold
le
v
el.
Depending
upon
intensity
v
alues
of
the
gi
v
en
image
g
amma
changes
according
to
the
condition
as
in
equation
(5).
F
or
high
intensity
v
alues
we
maintain
the
local
contrast
by
using
cubic
function.
Simultaneously
,
the
g
amma
correction
for
high
intensi
ty
images
mai
ntains
the
local
contrast
and
remo
v
es
the
color
cast
present
in
the
image.
The
a
1
;
a
2
;
a
3
and
a
4
mentioned
in
equation
(5)
are
constants
and
get
e
v
aluated
using
the
equation
(11),
(12),
(13)
and
(14).
In
order
to
determine
the
coef
ficients
a
1
;
a
2
;
a
3
and
a
4
,
the
follo
wing
conditions
are
imposed
I(t)
=a
1
t
3
+a
2
t
2
+a
3
t
+a
4
=
t
(6)
I
0
(t)
=
3a
1
t
2
+2a
2
t
+a
3
=
1
(7)
I(m)
=a
1
m
3
+a
2
m
2
+a
3
m
+a
4
=
sm
(8)
IJECE
V
ol.
7,
No.
5,
October
2017:
2581
–
2595
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISSN:
2088-8708
2583
I
0
(m)
=
3a
1
m
2
+2a
2
m
+a
3
=s
(9)
where
I’
is
the
first
deri
v
ati
v
e
of
I,
denotes
the
maximum
input
le
v
el,
and
denotes
the
di
f
ferential
coef
ficients
of
the
con
v
entional
knee
curv
e
at
the
maximum
input
le
v
el
are
gi
v
en
by
s
=
1
k
t
k
(10)
Where
k
denotes
the
intensity
le
v
el
at
knee
point
of
the
con
v
entional
knee
curv
e.
W
e
get
t
he
coef
ficient
v
alues
are
as
follo
ws
a
1
=
(s
1)t
+
(2
m
ms)
(t
m)
3
(11)
a
2
=
2(1
s)t
2
+(ms
+
2m
3)(t
+
m)
(t
m)
3
(12)
a
3
=
st
3
+(s
4)m
t
2
+(6
m
2ms)m
t
m
3
(t
m
)
3
(13)
a
4
=
(1
ms)t
3
+(ms
+
2m
3)m
t
2
(t
m)
3
(14)
Similarly
,
we
can
perform
the
same
correction
for
other
remaining
channels
3.
P
AR
TICLE
SW
ARM
OPTIMIZA
TION
The
PSO
algorithm
implementation
can
be
summarized
as
Step
1:
Initialize
all
particles
randomly
according
to
the
solution
space
satisfying
the
computational
load
or
iterations
required
to
obtain
the
optimum
solution
as
sho
wn
in
T
able
1
Step
2:
loop:
F
or
each
particle,
obtain
the
fitness
function
in
D
v
ariables
do:
Step
3:
Set
the
v
alue
as
the
maximum
v
alue
between
the
current
v
alue
and
e
xisting
v
alue.
Step
4:
Identify
the
particle
in
the
neighborhood
with
the
best
success
so
f
ar
,
and
assign
its
inde
x
to
the
v
ariable
t.
Step
5:
Update
the
particle
v
elocity
using
v
t+1
i
=W
t
:
v
t
i
+c
1
:
r
1
:
(pb
est
t
i
X
t
i
)+c
2
:
r
2
:
(gb
est
t
X
t
i
)
(15)
Where
r
1
and
r
2
are
random
v
alues
generated
in
the
range
[0,
1].
Step
6:
Update
the
particle
position
using
X
t+1
i
=
X
t
i
+v
t+1
i
(16)
Step
7:
If
a
criterion
(i.e.
usually
a
suf
ficiently
good
fitness
or
a
maximum
number
of
iterati
ons)
is
met,
then
terminate
the
loop.
T
able
1.
PSO
P
arameters
P
arameters
V
alues
Population
size
100
Max
iteration
200
W
m
a
x
0.9
W
m
i
n
0.4
c
1
2
c
2
2
4.
FITNESS
FUNCTION
FOR
OPTIMIZA
TION
In
this
w
ork,
the
mean
intensity
calculated
from
the
normalized
v
alues
of
R,
G,
and
B
channel
of
color
cast
test
image
has
to
be
preserv
ed
in
the
enhanced
color
image.
The
correction
f
actor
is
optimized
in
the
proposed
case
for
each
pix
el
v
alue.
The
intensity
v
alue
changes
for
each
pix
el
channel
in
the
image.
W
e
applied
this
f
actor
for
cast
image
which
is
not
based
on
the
an
y
assumption.
A
parameterized
transformation
function
be
defined
in
Intensity
Pr
eserving
Cast
Remo
val
in
Color
Ima
g
es
Using
P
article
Swarm
...
(Om
Pr
akash
V
erma)
Evaluation Warning : The document was created with Spire.PDF for Python.
2584
ISSN:
2088-8708
order
t
o
remo
v
e
the
color
cast
present
in
a
test
image
as
R
,
G
and
B
.
The
transformation
function
contains
the
parameters
which
is
a
real
v
alued
lying
between
0
and
10
such
that
the
mean
intensity
of
distorted
image
gets
preserv
ed.
The
amount
of
color
cast
adjustment
is
a
typical
task
by
using
con
v
entional
approach.
This
is
accomplished
by
e
v
aluating
the
optimal
v
alue
using
PSO
for
pix
el
v
alue
in
each
channel.
No
w
our
aim
is
to
find
out
the
optimized
set
of
real
v
alues
of
=
i
j
,i
and
j
ar
e
the
pix
el
locations
in
an
image
for
each
channel
by
using
PSO
which
produces
an
acceptable
output
as
per
the
fitness
function.
Here,
the
fitness
function
J
for
each
channel
is
proposed
as
J
=
I
(17)
Similarly
,
we
can
define
the
fitness
function
for
G
and
B
channels.
The
proposed
algorithm
firstly
initializes
P
number
of
particles.
This
means
that
the
posit
ion
v
ector
of
each
particle
X
has
one
component
of
.
Further
,
using
this
parameter
v
alue
in
each
generat
ion,
the
particle
remo
v
es
the
color
cast
using
the
intensity
transformation
function
defined
for
each
channel
as
R
.
T
ransformation
function
changes
the
v
alue
of
each
pix
el
in
the
test
image
according
to
the
parameter
v
alues.
The
v
alues
of
g
amma
modify
the
intensity
of
each
pix
el
and
also
preserv
e
the
intensity
of
image.
These
g
amma
v
alues
remo
v
es
color
cast
up
to
some
e
xtent
and
produced
a
number
of
color
corrected
images.
Fitness
v
alues
of
all
the
corrected
images
generated
by
all
the
p
a
rticles
are
calculated.
These
pbest
and
gbest
locations
are
gi
v
en
by
fitness
v
alues
according
to
the
fitness
function
defined
in
equation
(17).
The
is
the
best
solution
of
a
particular
particle
that
it
has
achie
v
ed
so
f
ar
,
it
is
also
referred
to
as
cogniti
v
e
component
which
update
their
beha
vior
only
as
per
t
h
e
ir
o
wn
e
xperience
and
another
best
v
alue
which
is
called
as
referred
as
social
component
are
e
xplained
by
equation
(15).
In
this
component,
each
indi
vidual
ignore
its
o
wn
e
xperience
and
update
their
beha
viour
according
to
the
pre
vious
best
particle
in
the
neighbourhood
of
the
group.
This
cogniti
v
e
component
combines
with
social
component
by
the
updating
formula
gi
v
en
in
equation
(15)
and
equation
(16)
and
calculates
the
ne
w
v
elocity
of
each
particle.
When
the
process
is
completed,
the
color
corrected
image
is
created
by
the
position
of
the
particles
as
it
pro
vides
the
maximum
fitness
v
alue.
Further
,
using
this
paramet
er
v
alue
in
each
generation,
the
particle
remo
v
es
the
color
cast
using
the
intensity
transformation
function
defined
for
each
channel
as
R(i
;
j)
Algorithm:
Color
correction
algorithm
using
PSO
1.
Input:
A
color
image
X
=
R,
G,
B
of
size
M
x
N
pix
els
2.
Obtain
the
normalized
v
alue
of
each
channel
R
,
G
,
B
using
equation
(1),
(2)
and
(3).
Compute
the
mean
intensity
v
alue
using
the
equation
(4).
3.
Define
PSO
parameters:
particles
P
n
,
iterations
itr
n
4.
Firstly
consider
the
red
channel
and
obtain
the
modified
red
channel
I
after
applying
the
transfer
function
using
equation
(5).
5.
Compute
the
set
of
v
alues
that
optimizes
the
fitness
function
gi
v
en
in
equation
(17)
using
PSO
technique,
at
each
pix
el
location
for
the
selected
channel
as
input
(i.e.
obtain
M
x
N
v
alues).
6.
Apply
the
resulted
g
amma
v
alues
and
obtain
the
enhanced
corrected
R
channel.
7.
Repeat
the
abo
v
e
mentioned
steps
for
G
and
B
channel.
8.
Obtain
the
enhanced
corrected
channels
i.e.
G
c
h
a
n
n
e
l
and
B
c
h
a
n
n
e
l
.
Output:
o
v
erall
enhanced
brightness
preserving
color
corrected
image
Y=
R
c
h
a
n
n
e
l
;
G
c
h
a
n
n
e
l
;
B
c
h
a
n
n
e
l
]
T
ransformation
function
changes
the
v
alue
of
each
pix
el
in
the
test
image
a
ccording
to
the
parameter
v
alues.
Fitness
v
alues
of
all
the
corrected
images
generated
by
all
the
particles
are
calculated.
These
and
locations
are
determined
according
to
t
he
fitness
function
defined
in
equation
(17).
When
the
process
is
complete
d
,
the
color
corrected
image
is
created
by
the
position
of
the
particles
as
it
pro
vides
the
maximum
fitness
v
alue.
5.
PERFORMANCE
MEASURE
CIELAB
metric
[19]
estimates
accurac
y
of
the
color
reproduction
in
comparison
to
the
original
when
analyzed
by
a
human
observ
er
.
The
CIELAB
metric
is
suitable
for
measuring
color
dif
fer
ence
of
lar
ge
uniform
color
tar
gets.
CIELAB
is
based
on
one
channel
for
Luminance
(L)
and
tw
o
color
channels
(a
and
b).
The
a-axis
starts
from
green
(-a)
to
red
(+a)
and
the
b-axi
s
starts
from
blue
(-b)
to
yello
w
(+b).
The
Luminance
(L)
starts
from
the
bottom
to
the
top
of
the
three-dimensional
model.
The
E
c
and
E
e
metric
are
gi
v
en
by
E
c
=
q
L
2
c
+a
2
c
+b
2
c
(18)
IJECE
V
ol.
7,
No.
5,
October
2017:
2581
–
2595
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISSN:
2088-8708
2585
where
L
c
=
L
cast
L
original
a
c
=a
cast
a
original
b
c
=b
cast
b
original
E
e
=
q
L
e
2
+a
e
2
+b
e
2
(19)
where
L
e
=
L
enhanced
L
original
a
e
=a
enhanced
a
original
b
e
=b
enhanced
b
original
P
arameters
L
c
;
a
c
and
b
c
is
the
dif
ference
in
the
cast
and
test
image
coordinates
of
L,
a,
and
b
of
CIELAB
color
space
and
L
e
;
a
e
and
b
e
are
the
coordinates
L,
a,
and
b
of
enhanced
image
in
CIELAB
col
or
space.
is
the
Euclidean
distance
for
measuring
the
dif
ference
between
colors.
Smaller
v
alue
of
indicates
that
the
enhanced
test
image
is
closer
to
the
reference
image.
6.
RESUL
TS
AND
DISCUSSIONS
The
proposed
method
has
been
successfully
implemented
using
MA
TLAB
7.10.
W
e
ha
v
e
been
tested
50
images
under
a
di
v
ersified
illumination
conditions
and
the
results
of
sample
images
(viz.
Building,
Stanford
T
o
wer
,
House,
Mandrill,
V
illage,
T
ree,
T
w
o
men,
Lena)
are
illustrated
in
the
paper
.
W
e
tested
the
proposed
algorithm
ag
ainst
gray
w
orld
assumption
and
Kw
ok
method
[17]
using
CIELAB
E
metric.
The
cast
images
and
images
obtained
from
the
Gray
w
orld
corrected
approach;
Kw
ok
method
and
the
proposed
method
ha
v
e
been
sho
wn
in
Figures
1-8.
The
reference
image
is
the
original
image
without
an
y
bad
illumination
ef
fect.
The
distorted
images
are
obtained
by
adding
color
cast
to
them.
The
distorted
images
are
corrected
by
using
gray
w
orld
approach,
Kw
ok
approach
and
the
proposed
corrected
approach.
The
image
distances
are
calculated
using
equation
(18),
equation
(19)
and
summarized
in
T
able
2.
In
Figures
1-8,
(a)
sho
ws
the
original
image;
(b),
(c)
and
(d)
sho
w
the
red,
green
and
blue
cast
of
an
image;
(a),
(e),
(f),
(g)
represent
the
gray
w
orld
corrected
image
of
(b),
(c)
and
(d)
respecti
v
ely;
(h),
(i
)
and
(j)
sho
w
the
result
of
the
Kw
ok
corrected
images
of
(b),
(c)
and
(d)
respecti
v
ely;
(k),
(l)
and
(m)
sho
w
the
result
of
the
proposed
approach
corrected
images
of
(b),
(c),
and
(d)
respecti
v
ely
.
Figure
1
sho
ws
the
cast
b
uilding
images,
the
cl
ouds
are
not
appea
ring
to
be
blui
sh
in
color
a
nd
grass
color
changes
in
appearance
from
the
original
or
reference
image
in
dif
ferent
cast
conditions.
When
we
apply
the
gray
w
orld
corrected
algorithm
and
Kw
ok
correction
technique
on
the
cast
b
uilding
images
then,
the
resulting
images
sho
wn
in
Figure
1(e),
(f)
and
(g)
become
dark
which
significantly
reduces
the
color
originality
The
enhanced
images
obtained
by
using
Kw
ok
method
sho
wn
in
Figure
1
(h),
(i)
and
(j),
depict
that
color
correction
has
been
achie
v
ed
to
some
e
xtent.
The
ef
fect
of
color
cast
still
remains
in
the
enhanced
images.
The
images
obtained
by
proposed
method
sho
w
that
the
blue
color
clouds
and
green
color
grass
resembles
the
visual
appearance
as
it
is
in
the
reference
image.
The
same
results
depict
from
the
v
alue
i.e.
0.0475,
0.0421
and
0.0521
for
red
cast,
green
cast
and
blue
cast
respecti
v
ely
(near
to
0)
as
mentioned
in
T
able
2
The
cast
Stanford
T
o
wer
image
in
Figure
2,
enhanced
by
gray
w
orld
algorithm
and
Kw
ok
method
lacks
a
good
visual
appearance
while
the
proposed
algorithm
remo
v
es
dark
er
portion
of
the
image.
This
ensures
that
the
proposed
method
has
good
perceptibility
with
the
reference
image.
Further
,
the
results
are
supported
by
v
alue
0.013,
0.019
and
0.019
for
red
cast,
green
cast
and
blue
cast
respecti
v
ely
.
In
the
cast
House
image
Figure
3,
visual
analysis
re
v
eals
that
the
image
enhanced
by
proposed
me
thod
is
better
than
other
tw
o
e
xisting
methods.
The
proposed
algorithm
maintains
the
originality
of
roof
color
a
n
d
tree
lea
v
es
as
compared
to
the
gray
w
orld
correct
ion
algorithm.
The
House
image
enhanced
by
Kw
ok
method
is
brighter
b
ut
color
cast
is
not
completely
remo
v
ed.The
same
results
are
confirmed
from
the
obtained
v
alue
of
CIELAB
color
space
metric
mentioned
in
T
able
2.
In
a
mandrill
gray
w
orld
enhanced
image
sho
wn
in
Figure
4,
the
color
of
the
nose
doesn’
t
appear
to
be
red
as
in
the
original
mandrill
i
mage
and
the
green
cast
are
remo
v
ed
to
some
e
xtent.
If
we
analyze
the
results
produced
by
Kw
ok
method
we
can
observ
e
that
the
y
are
close
to
the
reference
image
in
red
and
blue
cast
b
ut
not
with
the
green
cast
image.
Therefore,
the
proposed
method
has
acquired
the
originality
of
the
color
by
remo
ving
the
cast.
Similarly
,
in
Figures
5,
6,
7
and
8
the
good
perceptible
quality
enhanced
image
has
been
achie
v
ed
using
the
proposed
approach.
The
enhanced
images
obtained
from
the
proposed
algorithm
are
near
to
the
original
or
reference
image
whereas
the
gray
w
orld
correction
algorithm
does
not
gi
v
e
good
result
when
there
is
a
lar
ge
contrib
ution
of
one
color
.
The
quantitati
v
e
results
of
the
Figures
1-8
are
gi
v
en
in
T
able-2.
This
table
sho
ws
that
the
statistical
parameter
of
the
proposed
approach
pro
vides
the
best
performance
as
compared
with
other
methods.
Intensity
Pr
eserving
Cast
Remo
val
in
Color
Ima
g
es
Using
P
article
Swarm
...
(Om
Pr
akash
V
erma)
Evaluation Warning : The document was created with Spire.PDF for Python.
2586
ISSN:
2088-8708
T
able
2.
Comparati
v
e
results
for
grayw
orld,
K
o
wk
and
Proposed
method
(Distance
Metric)
Reference
image
T
est
image
Ec
(T
est
Image)
Ee
(Grayw
orld)
Ee
(K
o
wk
method)
Ee
(Proposed
approach)
Image-1(Building)
Red
cast
0.1704
0.1186
0.0761
0.0475
Green
cast
0.1868
0.1353
0.0875
0.0421
Blue
cast
0.2189
0.1694
0.0768
0.0521
Image-2(Stanford
T
o
wer)
Red
cast
0.1825
0.0987
0.0232
0.0131
Green
cast
0.2400
0.1649
0.0721
0.019
Blue
cast
0.2340
0.1271
0.0640
0.0190
Image-3(House)
Red
cast
0.1746
0.1113
0.5680
0.0131
Green
cast
0.2070
0.1186
0.1089
0.0182
Blue
cast
0.2332
0.1264
0.1504
0.0181
Image-4(Mandrill)
Red
cast
0.2039
0.1251
0.0611
0.0490
Green
cast
0.2461
0.1510
0.0746
0.0450
Blue
cast
0.2790
0.1349
0.0785
0.0490
Image-5(V
illage)
Red
cast
0.2050
0.1050
0.0725
0.0470
Green
cast
0.2227
0.1745
0.0426
0.0450
Blue
cast
0.2316
0.1499
0.0817
0.0630
Image-6(T
ree)
Red
cast
0.1742
0.1085
0.0391
0.0320
Green
cast
0.2050
0.1459
0.0450
0.0320
Blue
cast
0.2140
0.1552
0.0583
0.0410
Image-7(T
w
o
men)
Red
cast
0.1360
0.1086
0.0813
0.0540
Green
cast
0.1781
0.1212
0.0921
0.0600
Blue
cast
0.1832
0.1316
0.0881
0.0600
Image-8
(Lena)
Red
cast
0.2342
0.1212
0.0164
0.0121
Green
cast
0.3151
0.1768
0.0196
0.0113
Blue
cast
0.3210
0.1144
0.0150
0.0131
7.
CONCLUSION
The
paper
has
presented
g
amma
correction
approach
for
color
image
enhancement
as
well
as
preserving
the
mean
intensity
of
the
image.
The
particle
sw
arm
optimization
algorithm
is
used
to
obtain
an
optimal
g
amma
v
alue
by
pres
erv
ation
of
intensity
v
alue
and
maximizing
the
information
content
.
The
results
ha
v
e
sho
wn
that
the
proposed
approach
performs
better
than
the
gray
w
orld
approach
and
the
recent
Kw
ok
method
as
well.
In
addition,
the
proposed
method
remo
v
es
the
color
cast
completely
.
The
ef
fecti
v
eness
of
the
proposed
approach
is
quantitati
v
ely
measured
by
distance
metric
E
of
CIELAB
color
space.
REFERENCES
[1]
Gonzalez,
R.C.
and
W
oods,
R.E.,
Digital
image
processing
.
2002.
[2]
Zhu,
H.,
Chan,
F
.H.
and
Lam,
F
.K.,
”Image
contrast
enhancement
by
constrained
local
histogram
equalization,
”
Computer
V
ision
and
Image
Understanding,
,
v
ol.
73(2),
pp.
281-290,
1999.
[3]
Kim,
J.Y
.,
Kim,
L.S.
and
Hw
ang,
S.H.,
”An
adv
anced
contrast
enhancement
using
partially
o
v
erlapped
sub-
block
histogram
equalization.,
”
IEEE
T
ransactions
on
Circuits
and
Systems
for
V
ideo
T
echnology
,
,
v
ol.11(4),
pp.475-484,
2001.
[4]
Buchsbaum,
G.,
”
A
spatial
processor
model
for
object
colour
perception,
”
Journal
of
the
Franklin
institute,
,
v
ol.310(1),
pp.1-26,1980.
[5]
T
ang,
B.,
Sapiro,
G.
and
Caselles,
V
.,
”Color
image
enhancement
via
chromaticity
dif
fusion.
”
IEEE
T
ransactions
on
Image
Processing,
,
v
ol.10(5),
pp.701-707,
2001.
[6]
Kw
ok,
N.M.,
W
ang,
D.,
Jia,
X.,
Chen,
S.Y
.,
F
a
n
g,
G.
and
Ha,
Q.P
.,
”Gray
w
orld
based
color
correction
and
in-
tensity
preserv
ation
for
image
enhancement
,
”,
4th
IEEE
International
Congress
on
Image
and
Signal
Processing
(CISP),
2011
,
v
ol.2,
pp.
994-998,
October
,
2011.
[7]
F
arid,
H.,
”Blind
in
v
erse
g
amma
correction,
”
IEEE
T
ransactions
on
Image
Processing,
,
v
ol.10(10),
pp.1428-
1433,2001.
IJECE
V
ol.
7,
No.
5,
October
2017:
2581
–
2595
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISSN:
2088-8708
2587
[8]
Monobe,
Y
.,
Y
amashita,
H.,
K
urosa
w
a,
T
.
and
K
otera,
H..
”Dynamic
range
compression
preserving
local
image
contrast
for
digital
video
camera,
”
IEEE
T
ransactions
on
Consumer
Electronics,
,
v
ol.51(1),
pp.1-10,
2005.
[9]
Saitho.
F
,
”Image
contrast
enhancement
using
genetic
algorithm,
in
Proceedings
of
IEEE
SMC,
T
ok
yo,
Japan,
pp.
899-904,
1993.
[10]
Zhang,
C.
and
W
ang,
X..
,
”Global
and
local
contrast
enhancement
for
image
by
genetic
algorithm
and
w
a
v
elet
neural
netw
ork,
”
Neural
Information
Processing
,
pp.
910-919,
Springer
Berlin
Heidelber
g,
2006.
[11]
Hashemi,
S.,
Kiani,
S.,
Noroozi,
N.
and
Moghaddam,
M.E.,
”An
image
contrast
enhancement
method
based
on
genetic
algorithm,
”,
P
attern
Recognition
Letters,
v
ol.31(13),
pp.1816-1824,
2010.
[12]
Guan,
X.,
Jian,
S.,
Hongda,
P
.,
Zhiguo,
Z.
and
Haibin,
G.,
”An
image
enhancement
method
based
on
g
amma
correction,
”
IEEE
Second
International
Symposium
on
Computational
Intelligence
and
Des
ign
2009
ISCID’09,
,v
ol.1,
pp.
60-63,
2009.
[13]
Eberhart,
R.C.
and
K
ennedy
,
J.,
”A
ne
w
optimizer
using
particle
sw
arm
theory
,
”,
Proceedings
of
the
sixth
international
symposium
on
micro
machine
and
human
science,
,
v
ol.
1,
pp.
39-43),
1995.
[14]
W
ang,
Y
.Q.
and
Liu,
W
.Y
.,
,
”Application
of
sw
arm
intelligence
in
image
processing,
”
Journal
of
Computer
Applications,
,
v
ol.27(7),
pp.1647-1650,
2007.
[15]
Braik.
M,
Aladdin
.
A,
Sheta.
A.,
”Image
enhancement
particle
sw
a
rm
optimization,
”
Proceedings
of
W
orld
Congress
on
Engineering,
,v
ol.1,
2007.
[16]
Kw
ok,
N.M.,
Ha,
Q.P
.,
Liu,
D.
and
F
ang,
G.,
”Contrast
enhancement
and
intensity
preserv
ation
for
gray-le
v
el
images
using
multiobjecti
v
e
particle
sw
arm
optimization,
”,
IEEE
T
ransactions
on
Automation
Science
and
En-
gineering,
”
,
v
ol.6(1),
pp.145-155,
2009.
[17]
Kw
ok,
N.M.,
Shi,
H.Y
.,
Ha,
Q.P
.,
F
ang,
G.,
Chen,
S.Y
.
and
Jia,
X.,
2013.
Simultaneous
image
color
correction
and
enhancement
using
particle
sw
arm
optimization.
Engineering
Applications
of
Artificial
Intelligence,
26(10),
pp.2356-2371.
[18]
Gorai,
A.
and
Ghosh,
A.,
”Hue-preserving
color
image
enhancement
using
particle
sw
arm
optimization,
”,
IEEE
Recent
Adv
ances
in
Intelligent
Computational
Systems
(RAICS),
2011,
,
pp.
563-568,
September
2011.
[19]
Zhang,
X.
and
W
andell,
B.A.,
”A
spatial
e
xtension
of
CIELAB
for
digital
colorimage
reproduction,
”
Journal
of
the
Society
for
Information
Display
,
,v
ol.5(1),
pp.61-63,
1997.
BIOGRAPHY
OF
A
UTHORS
Om
prakash
V
erma
He
recei
v
ed
his
B.E.
de
gree
in
Electronics
and
Communication
Engineering
from
Mala
viya
National
Inst
itute
of
T
echnology
,
Jaipur
,
India,
M
.
T
ech.
de
gree
in
Communi-cation
and
Radar
Engineering
from
Indian
Institute
of
T
echnology
(IIT),
Delhi,
India
and
Ph.D.
de
gree
from
Delhi
Uni
v
ersity
.
From
1992
to
1998
he
w
as
Assistant
Professor
in
Department
of
ECE
at
Mala
viya
National
Institute
of
T
echnology
,
Jaipur
,
India.
He
joined
Department
of
Electronics
Communication
Engineering,
Delhi
T
echnological
Uni
v
ersity
(formerly
Delhi
Colle
ge
of
Engineer
-
ing)
as
Associate
Professor
in
1998.
Currently
,
he
is
Professor
and
Head,
Department
of
Informa-
tion
T
echnology
Delhi
T
echno-logical
Uni
v
ersity
,
Delhi
India.
He
is
also
the
author
of
more
than
30
publications
in
both
referred
journals
and
international
conferences.
He
has
guided
more
than
35
M.
T
ech.
students
for
their
thesis
and
presently
5
research
scholars
are
w
orking
under
his
su-
pervision
for
their
Ph.D.
He
has
authored
a
book
on
Digital
Signal
Processing
in
2003.
His
present
research
interest
includes:
Applied
soft
computing,
Artificial
intelligent,
Ev
olu-tionary
computing,
Image
Processing,
Digi
tal
signal
processing.
He
is
also
a
Principal
in
v
estig
ator
of
an
Infor
-mation
Security
Education
A
w
areness
project,
sponsored
by
Department
of
Information
T
echnology
,
Go
v-
ernment
of
India.
He
has
a
started
online
admission
process
for
B.
T
ech
admissions
at
DTU
in
2011.
He
is
a
re
gular
re
vie
wer
of
man
y
International
Journals
lik
e
IEEE
transaction,
Else
vier
,
Springer
etc.
He
has
acted
as
program
committee
member
and
chaired
session
for
man
y
conferences.
Nitin
Sharma
He
recei
v
ed
his
B.T
ech.
de
gree
in
Electronics
and
Communication
Engineering
from
UPTU,
India
in
2005
and
M.T
ech
de
gre
e
in
Electronics
and
Communication
Engineering
from
MMMEC
Gorakhpur
,
UPTU,
India
in
2007.
Currently
he
is
w
orking
as
an
Assistant
Professor
in
Electronics
and
Communication
Engineering
Department
of
MAIT
,
Rohini,
De
lhi,
India.
His
re-
search
interests
are
in
the
area
of
Image
Processing,
color
image
enhancement
and
soft
computing.
Intensity
Pr
eserving
Cast
Remo
val
in
Color
Ima
g
es
Using
P
article
Swarm
...
(Om
Pr
akash
V
erma)
Evaluation Warning : The document was created with Spire.PDF for Python.
2588
ISSN:
2088-8708
(a)
Reference
image
(b)
Red
cast
(c)
Green
cast
(d)
Blue
cast
(e)
Gray
corrected
Red
cast
(f)
Gray
corrected
Green
cast
(g)
Gray
corrected
Blue
cast
(h)
Kw
ok
corrected
Red
cast
(i)
Kw
ok
corrected
Green
cast
(j)
Kw
ok
corrected
Blue
cast
(k)
Proposed
approach
corrected
Red
cast
(l)
Proposed
approach
corrected
Green
cast
(m)
Proposed
approach
corrected
Blue
cast
Figure
1.
Image-1:Building
IJECE
V
ol.
7,
No.
5,
October
2017:
2581
–
2595
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISSN:
2088-8708
2589
(a)
Reference
image
(b)
Red
cast
(c)
Green
cast
(d)
Blue
cast
(e)
Gray
corrected
Red
cast
(f)
Gray
corrected
Green
cast
(g)
Gray
corrected
Blue
cast
(h)
Kw
ok
corrected
Red
cast
(i)
Kw
ok
corrected
Green
cast
(j)
Kw
ok
corrected
Blue
cast
(k)
Proposed
approach
corrected
Red
cast
(l)
Proposed
approach
corrected
Green
cast
(m)
Proposed
approach
corrected
Blue
cast
Figure
2.
Image-2:T
o
wer
Intensity
Pr
eserving
Cast
Remo
val
in
Color
Ima
g
es
Using
P
article
Swarm
...
(Om
Pr
akash
V
erma)
Evaluation Warning : The document was created with Spire.PDF for Python.
2590
ISSN:
2088-8708
(a)
Reference
image
(b)
Red
cast
(c)
Green
cast
(d)
Blue
cast
(e)
Gray
corrected
Red
cast
(f)
Gray
corrected
Green
cast
(g)
Gray
corrected
Blue
cast
(h)
Kw
ok
corrected
Red
cast
(i)
Kw
ok
corrected
Green
cast
(j)
Kw
ok
corrected
Blue
cast
(k)
Proposed
approach
corrected
Red
cast
(l)
Proposed
approach
corrected
Green
cast
(m)
Proposed
approach
corrected
Blue
cast
Figure
3.
Image-3:house
IJECE
V
ol.
7,
No.
5,
October
2017:
2581
–
2595
Evaluation Warning : The document was created with Spire.PDF for Python.