TELKOM
NIKA
, Vol.11, No
.1, Janua
ry 2013, pp. 362
~37
0
ISSN: 2302-4
046
362
Re
cei
v
ed O
c
t
ober 3, 20
12;
Revi
se
d No
vem
ber 20, 20
12; Accepted
De
cem
ber 1
0
,
2012
Single Image Haze Removal Method for Inland River
Zhong
y
i
Hu*
1,2
, Qiu Liu
1,
3
1,*
Intellige
n
t T
r
ansp
o
rt S
y
ste
m
s Researc
h
Center,
W
uha
n
Universit
y
of Techn
o
lo
g
y
, W
uhan, Hu
be
i
430
07
0, P.R.
Chin
a.
2
Intellig
ent Info
rmation S
y
ste
m
s Institute, Wenzh
ou
U
n
iver
sit
y
, Wenzh
ou,
Z
hejia
ng 3
250
35, P.R.Chin
a
3
Intellig
ent T
r
a
n
sport S
y
stem
s Researc
h
Ce
nter, W
uhan U
n
iversit
y
of T
e
chno
log
y
,
W
uhan, Hu
bei
430
07
0, P.R.
Chin
a.
*Corres
p
o
ndi
n
g
author, e-ma
il:
huju
n
y
i@
16
3.com, Qliu20
0
0
@1
63.com
A
b
st
r
a
ct
Due
to
envir
on
me
ntal
p
o
ll
utio
n, the
cli
m
ate i
s
w
o
rseni
ng. T
he fo
g
days
up
to 6
0
%
of th
e
year
i
n
i
n
l
a
n
d
ce
rta
i
n
se
gm
en
ts, wh
i
c
h
i
t
ha
s se
ri
ousl
y
a
ffe
cte
d
th
e ma
ri
ne
el
e
c
tron
i
c
crui
se
n
o
r
m
a
l
op
e
r
a
t
i
o
n
a
nd
navi
gatio
n safe
ty. According t
o
the in
lan
d
vid
eo i
m
a
ge
b
e
co
mes
gray a
nd l
a
ck of visib
ility
in foggy w
eat
h
e
r
cond
itions,
and
in or
der to re
mov
e
the
ha
z
e
to get
a cl
ear
imag
e col
o
r an
d cont
o
u
r, this
pap
er pres
ents
a
meth
od
bas
ed
on Jo
nes Exte
nsio
n Matrix
a
nd the
Dark
Ch
ann
el Pri
o
r. F
i
r
s
t, w
e
obtain th
e li
ght i
n
tensity
i
n
the at
mos
pher
e an
d th
e esti
mate
d co
nce
n
tration
of the
h
a
z
e
by
usi
ng
Dark C
han
ne
l
Prior, an
d vi
a
usin
g
the Jon
e
s Extensio
n Matrix a
nd the p
a
ra
me
ters of St
okes'
Law
to eli
m
in
ate part
of the
scattered li
ght.
At
last, w
e
have compl
e
ted the f
unctio
n
of i
m
ag
e deh
a
z
i
n
g
by
brig
htness a
d
ju
stment factor b
a
sed o
n
N pix
e
l
s
in the fi
eld
of
step bri
ghtn
e
s
s
and
i
m
prov
e
the bri
ghtn
e
s
s
base
d
o
n
R
e
tinex Pr
inci
pl
e for the rec
o
v
e
red
imag
e. Experi
m
e
n
tal res
u
lts show
this algo
rithm i
m
prov
es
scenery visu
al
effect in condi
tion of ha
z
e
. It i
s
provi
ded a cl
ea
r video i
m
age f
o
r the marin
e
e
l
ectron
ic cruise
in the foggy d
a
y.
Key
w
ords
:
Jones Extensio
n Matrix, Stoke
s
' Law, Dark Chann
el Prior, Retinex, Marine
Electronic Crui
se.
Copyrig
h
t
©
2013
Univer
sitas Ahmad
Dahlan. All rights res
e
rv
ed.
1. Introduc
tion
In the variety of modes
of transporta
tion,
water transportatio
n
is one of th
e most
importa
nt tra
n
sp
ortation,
and the i
n
lan
d
wate
r tr
a
n
s
po
rt is the
most imp
o
rta
n
t comp
one
n
t
of
water t
r
an
sp
o
r
t [1]. Due to
environ
menta
l
pollution, th
e climate i
s
worsenin
g
. The
fog days u
p
to
60% of the
year in inl
a
n
d
ce
rtain
se
gments [2],
whi
c
h it ha
s seri
ou
sly affected th
e m
a
rine
electroni
c cru
i
se no
rmal o
peratio
n and
navigat
ion safety. Although many of the scien
c
e a
nd
techn
o
logy
workers
carrie
d out exten
s
i
v
e re
se
a
r
ch
on the fog i
m
age
cla
r
ity, and they h
a
ve
alrea
d
y a
c
hi
eved some
succe
ss, b
u
t i
n
land
river fog ima
ge
re
storatio
n p
r
o
b
lem i
s
still
one
vacan
c
y of
th
e do
me
stic
a
nd inte
rn
atio
nal
re
sea
r
ch, be
comi
ng
a
major p
r
oble
m
in th
e fiel
d
of
comp
uter visi
on until no
w.
In re
cent y
ears, mo
re
and mo
re
d
o
mest
i
c
an
d
foreign
re
search
schol
ars are
increa
singly
con
c
e
r
ne
d a
bout the
ato
m
ization
re
covery of the
deg
rade
d i
m
age, a
nd t
hey
prop
osed a lot of fog alg
o
rithm. At recent, there a
r
e tow majo
r catego
rie
s
deha
zing met
hod
s
inclu
d
ing i
m
a
ge e
nhan
ce
m
ent an
d ima
g
e
re
sto
r
ation
method. T
he
image
enh
an
ceme
nt meth
od
can
effectivel
y improve th
e fog ima
ge
contrast,
o
u
tstanding im
ag
e detail, to i
m
prove th
e v
i
sual
effect of the image. Such
as Rui Yibin [
3
]
Acco
rdin
g to the Retinex theory and
MSR algorith
m
,
norm
a
l inte
rception
stret
c
h
of foggy ima
ge p
r
o
c
e
ssi
n
g
. And
RUS
SO F [4] u
s
in
g wavelet mu
lti-
scale an
alysi
s
of the detail
s
of the fog in Figur
e e
quali
z
ation to fog. Method for im
age re
sto
r
atio
n
throug
h the establi
s
hm
en
t of the fog image de
grad
ation model i
n
vers
i
on de
g
r
adatio
n process
has b
een cl
e
a
r fog-free im
age, and its e
ffect to
fog method better t
han the imag
e enhan
ce
me
nt.
Tan [5] by
expandi
ng the
l
o
cal
co
ntra
st
of the resto
r
e
d
imag
e to a
c
hieve the
effect of fog. F
a
ttal
[6] assume t
hat the propa
gation of light
and
scene
t
a
rget
su
rface
sha
d
ing p
a
rt
is lo
cal an
d
not
related
to th
e premi
s
e, it is e
s
timate
d
that
out of t
he
scene i
r
radian
ce, a
n
d
thus de
rive
the
transmissivity map to
re
co
ver the im
age
. He [7] fou
n
d
that the d
a
rk ch
annel
pri
o
r law to
find th
e
transmissio
n
rate to
a
c
hie
v
e the ima
g
e
to fog. T
he
Li an
d Li
u [8]
ba
sed
on
th
e da
rk
chan
n
e
l
prio
r an
d th
e wavel
e
t co
efficients of correlati
on to
determi
ne th
e irregul
ar
re
gion, an
d fin
a
lly
resto
r
e
d
imag
e map and i
r
regula
r
area
s of the
transmi
ttance to obta
i
n the desi
r
ed
fog effect.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Single Im
age Ha
ze Rem
o
val Method for
Inland Ri
ve
r (Zhong
yi Hu)
363
This p
ape
r is inspi
r
ed
and
pre
s
ent
s a n
e
w
imp
r
oved
method to d
e
f
og: Acco
rdi
n
g to the
inland vid
eo i
m
age
be
com
e
s g
r
ay a
nd l
a
ck of visi
b
ility in foggy we
ather
co
nditio
n
s, an
d in
order
to remove th
e haze to get
a clea
r imag
e colo
r an
d contour, thi
s
p
aper
pre
s
e
n
ts a method b
a
s
ed
on
Jon
e
s Ext
ensi
on M
a
trix
and
the
Da
rk Ch
annel
Prio
r. First, we
o
b
tain the
light
inten
s
ity in the
atmosp
he
re
and th
e e
s
tim
a
ted
con
c
e
n
tration of
t
he
h
a
ze
by u
s
in
g
Dark Chann
e
l
Prio
r, an
d vi
a
usin
g the Jo
nes Exten
s
io
n Matrix and
the param
et
ers
of Stokes' Law to elim
inate part of
the
scattered lig
ht. At last,
we have
co
mpleted
the
function of i
m
age de
ha
zi
ng by bright
ness
adju
s
tment fa
ctor ba
se
d o
n
N pixel
s
i
n
the fiel
d
of
step b
r
ig
htne
ss and
imp
r
ove
the
brig
htne
ss
based on
Ret
i
nex Princi
ple
for the recovered im
age.
The metho
d
is rea
s
o
nabl
e in physi
cs, an
d it
is even a
b
le t
o
deal
with ta
rget that is v
e
ry fa
r a
w
ay
from came
ra
in heavy fog. Mean
while, t
h
is
method de
pe
nds n
e
ither o
n
two or mo
re input im
age
s of different
polari
z
atio
n d
i
rectio
ns, no
r
on
the possi
bility that the light
transmi
ssion function has
l
a
rger
vari
ance
or
shadow
exists on targe
t
surfa
c
e. It ca
n avoid halo
effects
by usi
ng the media
n
operation.
2. Jones Extension Ma
trix
Polari
zation i
s
a very imp
o
r
tant co
ncept
in
physi
cal o
p
tics,
we can
use va
riou
s
ways to
descri
be pol
a
r
ize
d
light an
d polari
z
atio
n
device
s
,
su
ch as matrix method, the index functio
n
method
and
bond
with th
e
ball m
e
thod,
esp
e
ci
ally,
the way to u
s
e
the Jone
s m
a
trix (Jone
s) a
n
d
Muelle
r mat
r
i
x
(Muell
e
r) re
pre
s
ent t
he
p
o
lari
zation
de
vices ha
s
a v
e
ry go
od
effe
ct. Jo
ne
s m
a
trix
and
M
uelle
r matrix
have
both simila
rities and
di
fferences, th
e t
w
o m
a
trix b
e
tween
the li
g
h
t
wave
s su
perpositio
n and t
he pha
se info
rmation of
the
operation is
different, the forme
r
doe
s n
o
t
kee
p
ph
ase
operation inf
o
rmatio
n, lig
hting waves
of coh
e
rent
sup
e
rp
ositio
n
,
while the l
a
tter
kee
p
p
h
a
s
e
operation i
n
formatio
n, co
here
n
t light
waves of supe
rpo
s
ition,
th
e
r
efore
we
ch
oose
Muelle
r matri
x
to partici
p
a
te in elimi
n
ating the
scattering
ope
rations.
Jon
e
s matrix incl
u
des
hori
z
ontal li
n
e
up pa
rtial d
e
vice, ve
rtical
line up p
a
rtia
l device, +
45
lines
up p
a
rtial devi
c
e,
-45
line
s
u
p
p
a
rtial d
e
vice,
1/4 wave pl
ate (ve
r
tical
f
a
st axi
s
), 1/4
wave
pl
ate
(level fast axi
s
),
dextral ro
und
ed partial d
e
vice an
d left-la
teral ro
und
ed
partial device, as is sho
w
n in table 1.
Table 1. Co
mmon Jone
s
Matrix
Op
tical
C
o
m
p
o
n
ent
Jones
matri
x
Op
tical
C
o
m
p
o
n
ent
Jones
matri
x
Horizontal
Pol
a
ri
z
e
r
1/4 Retar
dation
Sheet (Vertical)
Vertical
Pol
a
ri
z
e
r
1/4 Retar
dation
Sheet(Ho
r
izontal)
+45
o
Pol
a
ri
z
e
r
Right-Hand
ed
Circle Rotation
Pol
a
ri
z
e
r
-45
o
Pol
a
ri
z
e
r
Left-Han
ded
Circle
Pol
a
ri
z
e
r
Due
to the
m
i
st in th
e pi
ct
ure,
white
bal
anc
e
can
be
simplified
a
s
clo
s
e to
the
averag
e
image. F
o
r
some
com
p
lex
image, th
e
colo
r of th
e i
m
age
ch
ang
es,
white
bal
ance i
s
e
qua
l to
image’
s lo
cal
avera
ge a
p
p
roximately [
9
, 10]. To
offset the
scattering, thi
s
pa
per ta
ke
up t
he
hori
z
ontal a
n
d
vertical pol
arizer in the
matrix operations.
0
0
0
1
M
(1)
1
0
0
0
M
(2)
0
0
0
1
10
0
i
00
01
10
0
i
11
11
1
1
1
2
i
i
11
11
1
1
1
2
i
i
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 362 – 3
7
0
364
Equation (1
)
mean
s
th
e hori
z
ontal
po
la
rization
co
mpone
nts, e
quation
(2
)
mean
s vertical polari
z
atio
n comp
one
nts,
and expa
n
d
s the matrix
effectively.
(3)
(4)
In equation
(3) an
d (4
), s
mean
s scatte
ring
coeffici
e
n
t, and the value of it is d
e
cid
ed
by the scatte
ring inte
nsity, that is, by the atmo
sphe
re mist con
c
e
n
tration, exp
e
rime
nts
sho
w
ed
that, s rang
e
s
from 0
-
0.1,
distingui
sh l
e
vel is
less t
han 0.01. Stoke
s
La
w [1
1] (Stoke
s L
a
w,
1845
) ca
n b
e
applied to
visible light intensity,
if
let natural li
ght go throu
gh the pola
r
i
z
ed
element
s, its light intensity:
(5)
Among them,
stands for t
he inci
dent light intens
ity, while sta
n
d
s
for the intensity of
light after pol
arized.
Then
and
are
sub
s
tituted into the equati
on (5
), it will get the equati
on (6
).
(6)
3. The Deh
a
zing Method
Bas
e
d on Jo
nes Exte
nsi
on Matrix an
d Dark
Chan
nel Prior
3.1. Backg
ro
und
Re
sea
r
ch areas in the
ri
ver to fog, only need
to con
s
id
er the
transpo
rt pro
pertie
s
of
visible lig
ht in
the atmo
sph
e
re, research
ers can
ign
o
re the light
wa
ve freque
ncy.
Such
as fig
u
r
e
1. The li
ght e
n
tering
the
camera in
clud
es t
w
o
part
s
, one
is th
e h
a
ze
scatteri
n
g
form
ation
,
and the othe
r is dire
ct exposu
r
e
of the target obj
ect formatio
n
[12].
Figure 1. Inland River F
o
g
g
y Image
In com
puter
vision field
s
,
the model
wi
dely
used to
descri
be the
formation
of a ha
ze
image is e
q
u
a
tion (7
).
(7
)
M
M
0
0
1
s
M
1
0
0
s
M
M
0
I
I
0
I
M
M
M
I
I
M
I
I
0
0
S
I
R
I
))
(
1
(
)
(
)
(
x
t
A
x
t
L
x
I
object
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Single Im
age Ha
ze Rem
o
val Method for
Inland Ri
ve
r (Zhong
yi Hu)
365
Whe
r
e
is the light entering
the camera,
is the distan
ce of the ca
mera an
d the
obje
c
t,
is the light of the target obj
ect
formatio
n
,
is the atmosp
he
re m
edium
transmissio
n descri
b
ing th
e portio
n
of the light
that is not scatte
red and
rea
c
h
e
s the
came
ra,
and
is the
gl
obal atmo
sp
h
e
ric light. Accordin
g to
Fig
u
re
1, the eq
uation
(7)
ca
n be
re
written
as eq
uation (8).
(8)
Whe
r
e
is the reflected ligh
t
of the
target
object format
ion, it is actually the object
light intensity
by the atmosph
e
re (fog atmosp
he
re) into
the
ob
se
rved st
rength
of the came
ra,
is the scattered light of the atmosp
h
e
re. So we wo
uld
get the equati
on (9
).
(9)
3.2. The Principle of Dehazin
g
Method Ba
sed
on Jones E
x
ten
s
ion Ma
trix and
Da
rk
Channel Prior
The prin
cipl
e
of the algorithm sho
w
n in
Fi
gure 2, it would proc
ess the image by Jone
s
Extension M
a
trix and Da
rk Ch
ann
el Prior be
cau
s
e
of the inland
foggy im
age
characte
ri
stics.
First, it
coul
d obtai
n the
scattere
d li
ght ap
pr
oxim
ate strength
by J
one
s Extension M
a
trix
operation
s
, a
nd then
calculate the
refl
ected
light
i
n
tensity. Second, it could
cal
c
ulate
th
e
atmosp
he
ric
optical
tran
sf
er fu
nction
a
c
cording
to the
Da
rk
Chan
n
e
l Prio
r, a
nd t
hen im
prove t
h
e
brightn
e
ss of the rest
ore
d
image; at last
it would obtai
n the desi
r
ed
deha
zin
g
effect.
Figure 2.
The
Principl
e of Deh
a
zi
ng Me
thod Base
d o
n
Jon
e
s Extensio
n Matrix and Dark
Cha
nnel Prio
r
Equation (10
)
cal
c
ulated
by (6), (7) a
nd (9), it
coul
d eli
m
inate some
of the scattered light
and get a ne
w image.
)
(
x
I
x
object
L
)
(
x
t
A
)
(
)
(
)
(
x
I
x
I
x
I
S
R
)
(
x
I
R
)
(
x
I
S
)
(
)
(
x
t
L
x
I
object
R
Inland Rive
r Ha
zing Imag
e
Jon
e
s
Extension
Matrix
Dar
k
Cha
nnel
Prior
Scattere
d Lig
h
t
Intens
ity
Atmos
p
heric
Tran
sfe
r
Fun
c
tion
Re
covered Image
Brightne
ss
Improving
Inland Rive
r Deh
a
zi
ng Image
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 362 – 3
7
0
366
(10
)
Obviou
sly, the inten
s
ity of
light ente
r
ing
the di
rect i
r
ra
diation of th
e
target o
b
je
ct
can
be
expre
s
sed a
s
equation (11
)
.
(11
)
Refer
e
n
c
e
7
g
i
ves the
app
roximate eq
ua
tion of the
atmosp
he
ric
op
tical tra
n
sfe
r
f
unctio
n
,
su
ch a
s
the type (12
)
as
shown.
(12
)
Actually,
is t
he d
a
rk
cha
n
nel of th
e o
r
i
g
inal
ha
zing i
m
age
.
It directly pro
v
ides the e
s
timation of t
he transmissio
n function for at
mosp
he
ric.
Becau
s
e of th
e worse
n
ing
of the global envir
on
ment, even throu
gh in the sunny
weath
e
r
of meteorology sense, the air
will al
ways have
so
me scatteri
ng parti
cles
. Therefore, when we
observe the d
i
stan
ce obj
ect
,
the haze will
always exi
s
t, and influen
ce peopl
e’s visual habit. More
importa
nce is that the haze
is an importa
nt clue to
the [12, 13] that the huma
n
eyes jud
ge dept
h
of field. Thi
s
phen
omen
on
is called
air
perspe
c
ti
ve.
Therefore,
in
order to
me
et peo
ple
visual
deman
d, the
algorith
m
of this p
ape
r i
s
n
o
t com
p
letely
to deha
ze,
b
u
t sele
ctively reserve
a sm
all
amount of h
a
ze for di
sta
n
t sce
ne, the deha
ze
d image after t
h
is will loo
k
more n
a
tural
.
In
equatio
n (12) by intro
d
u
c
i
ng a
con
s
tan
t
type param
eters,
, the equation
(1
2)
can
be
rewritten for e
quation (13
)
.
(13
)
It is very similar between t
he brig
htne
ss in
the cro
s
s of the sky an
d the river di
stan
ce
and the atm
o
sp
heri
c
lig
ht. In this pap
er, it wa
s ta
ken i
n
the e
x
perime
n
tal cal
c
ulatio
n of the
estimate
s. In fact, the estimate is relia
b
l
e bec
a
u
se of the sky and
the river cro
s
sover from the
came
ra
at infinity. So, the value of atmo
sph
e
ri
c
opti
c
al tran
sfer f
u
nction
tend
s
to 0. That is
the equatio
n (14).
(14
)
It would get
.
3.3. Brightn
e
ss Improv
ing Base
d on Retin
e
x
3.3.1. Backg
round of
Retinex
Retinex theo
ry is prop
ose
d
by the Lan
d,
who in the
1970
s of the last century model
based on the
colo
r and b
r
i
ghtne
ss of co
lor const
anc
y
ideas o
b
je
cts, huma
n
visual perce
ptio
n to
a point of lig
ht doe
s not d
epen
d on th
e
absolute lig
h
t
values, and
the su
rroun
d
i
nglight value
is
related
to th
a
t
huma
n
p
e
rception
of the
colo
r
of
obj
ects de
pen
ds o
n
the
surfa
c
e
ch
aracte
risti
cs
of the refle
c
ted light, and
has
nothing t
o
do with t
h
e
incide
nt [14, 15]. The theo
ry assum
e
s t
hat
the light i
n
te
nsity of a
im
age
by the
reflectio
n
co
mpone
nt
and
the radiation
comp
one
nt
,
su
ch a
s
equ
a
t
ion (15
)
.
M
x
I
M
x
I
x
t
L
object
)
(
)
(
)
(
)
(
)
(
)
(
x
t
M
x
I
M
x
I
L
object
))
)
(
(
min
(
min
1
)
(
)
(
c
c
x
y
c
A
y
I
x
t
))
)
(
(
min
(
min
)
(
c
c
x
y
c
A
y
I
c
c
A
y
I
)
(
)
1
,
0
(
))
)
(
(
min
(
min
1
)
(
)
(
c
c
x
y
c
A
y
I
x
t
)
(
x
t
0
))
)
(
(
min
(
min
1
)
(
)
(
c
c
x
y
c
A
y
I
x
t
c
c
A
I
)
,
(
y
x
I
)
,
(
y
x
R
)
,
(
y
x
S
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Single Im
age Ha
ze Rem
o
val Method for
Inland Ri
ve
r (Zhong
yi Hu)
367
(15
)
The ge
ne
ral
pro
c
e
ss
of the Retinex al
g
o
rith
m
sho
w
n
in Figure 3, then a
c
cordin
g to the
different meth
ods to e
s
tima
te the brightn
e
ss
of the image to get a d
i
fferent enha
n
c
eme
n
t.
Figure 3. The
General Pro
c
e
ss of Retin
e
x Algorithm
It puts the lo
garithm
ope
rators on
equ
ation (
15)
bot
h sid
e
s. An
d
it can
be o
b
tained
by
equatio
n (16
)
.
(16
)
Let
, then it c
an be obtain
ed
, at last
it can be obta
i
ned eq
uation
(17).
(17
)
Let
, and It puts the sou
r
ce image
smo
o
thing filter o
perato
r
s
and log
a
rithm
operators, th
en put
s the result to
in eq
uation (17), a
t
last puts th
e
antilog
operators o
n
the ne
w eq
uation (17
)
a
nd get Sin
g
l
e
-Scale
Reti
nex Algorith
m
[15]. Such
as
equatio
n (18
)
.
(18
)
Whe
r
e
is the
enhan
ce
d re
flection comp
onent,
is the
core of Gau
ss,
is the n
o
rm
alizatio
n co
n
s
tants,
is a
con
s
t
ant to
control the
scale of the G
a
ussian filter
rang
e. Experi
m
ents
sho
w
that c in
, the gr
ay dynami
c
co
ntra
st en
han
ced to a
c
hieve
better bala
n
ce, there is no
obviou
s
halo
effect.
3.3.2. V Enh
a
nced
for Re
store
d
Image
HSV col
o
r
sp
ace, al
so
kno
w
n a
s
HSB
color
spa
c
e, relative RGB
colo
r spa
c
e,
a more
accurate pe
rceptio
n of co
lor an
d bri
g
h
t
ness on th
e
conta
c
t and
remai
n
in th
e cal
c
ulatio
n
of
simple, typically obtained
inland fog i
m
age
s ar
e converted f
r
o
m
RGB spa
c
e to HSV
color
spa
c
e[1
6
]. Th
e V co
mpo
n
e
n
t is
reflecte
d
in the b
r
ig
htness of th
e i
m
age i
n
form
ation in th
e HSV
spa
c
e, its v
a
lue dete
r
mi
nes the
co
rresp
ondi
ng pi
xel shadi
ng
to a reaso
nable rang
e
o
f
enha
nced, can imp
r
ove t
he ove
r
all
bri
ghtne
ss of th
e imag
e afte
r
the re
store
of
the river fog
image.
It gets the
riv
e
r fo
g ima
ge
recover the
resu
lt
s from e
quation
(1
1)
conve
r
ted f
r
o
m
RGB sp
ace to HSV spa
c
e
,
filtering with a Gaussia
n
template of its V comp
one
nt. This pape
r
introdu
ce
s th
e function a
s
the imag
e
brightne
ss
adju
s
tment factor
, the functio
n
step
)
,
(
)
,
(
)
(
y
x
S
y
x
R
x
I
)
,
(
log
)
,
(
log
)
,
(
log
y
x
S
y
x
R
y
x
I
)
,
(
log
)
,
(
)
,
(
log
)
,
(
)
,
(
log
)
,
(
y
x
S
y
x
s
y
x
R
y
x
r
y
x
I
y
x
i
)
,
(
)
,
(
)
,
(
y
x
s
y
x
r
y
x
i
)
,
(
)
,
(
)
,
(
y
x
s
y
x
i
y
x
r
2
2
2
)
,
(
c
y
x
e
y
x
G
)
,
(
y
x
I
)
,
(
y
x
s
)))
,
(
*
)
,
(
log(
)
,
(
exp(
)
,
(
'
y
x
G
y
x
S
y
x
i
y
x
R
)
,
(
'
y
x
R
)
,
(
y
x
G
c
]
100
,
80
[
object
L
+
S(x,y
)
R(x,y
)
r(x,y
)
I(x,y
)
Lo
g
Im
age
Brightness
Es
tim
ates
Ex
p
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 362 – 3
7
0
368
brightn
e
ss of
the pixel wh
e
r
e V co
mpo
n
ent of t
he N
neigh
bori
ng p
i
xels to invad
e
the bri
ghtn
e
ss
variation
s
in
different dire
ction
s
, as
sh
own in
Fi
gu
re 3, whi
c
h is equivalent t
o
a neig
hbo
rhood
whe
r
e pixel
s
by pixels
Gau
ssi
an filter
windo
w,
t
he bri
ghtne
ss adj
ustme
n
t factor
su
ch
as
equatio
n (19
)
sho
w
s, then
the Gau
ssi
an
approve
d
the
re-d
efined a
s
equation (20
)
.
Figure 4. N Neighb
orh
ood
Pixel Step Brightne
ss
Diag
ram
(19
)
(20
)
Whe
r
e is a lin
ear regul
ator,
is the scal
e
factor, in this
article exp
e
ri
ments
and
.
4. Experimental Re
sult
It shoot a set
of foggy expe
rimental im
ag
e us
in
g the Sony DSC-T5
digital ca
mera in the
Yangtze
Riv
e
r Ri
pari
an.
Acco
rdin
g to deha
zin
g
method p
r
op
ose
d
in this pape
r, the first
combi
nation
of Jo
ne
s Extensi
on M
a
trix
and
Dark
Ch
annel
Prio
ri t
o
defo
g
, an
d
then th
rou
gh
the
Retinex pri
n
ciple to impro
v
e the brightness, sho
w
n
result
s in Figure 6. Wh
ere (a) is o
r
igi
nal
sou
r
ce imag
e
,
(b) is d
eha
zing image. It comp
are
the gray hi
stogra
m
of original
sou
r
ce imag
e
and g
r
ay hist
ogra
m
of de
hazi
ng imag
e
,
shown re
su
lts in Figure
7. Whe
r
e (a) is histo
g
ram
of
origin
al source image,
(b) i
s
hi
stogram o
f
deha
zing im
age. The
gra
y
histogram o
f
original
so
urce
image is
bet
wee
n
150 a
n
d
200, app
arently beca
u
se
of haze, an
d the goal of
scene i
s
ha
zy and
white. The g
r
ay histogram
of dehazi
ng
image is
bet
wee
n
0 and
250, it get a clea
r structu
r
es
and colors im
age, also accord with visua
l
rule.
N
i
i
V
V
1
0
2
2
2
2
2
2
1
0
)
,
(
c
y
x
N
i
i
c
y
x
e
V
V
k
e
k
y
x
G
c
8
k
80
c
1
N
N-
1
2
0
……
3
4
5
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Single Im
age Ha
ze Rem
o
val Method for
Inland Ri
ve
r (Zhong
yi Hu)
369
(a)
(b)
Figure 5.
Experimental Results
(a)
(b)
Figure 6.
Gra
y
Histogram
Comp
ari
s
o
n
of Original So
urce and
Deh
a
zin
g
Image
5. Conclusio
n
The pu
rpo
s
e
of this pape
r is to reali
z
e th
e
image to de
haze, prop
osed a very sim
p
le but
effective to d
eha
ze
algo
rithm for si
ngle
inland
foggy i
m
age. It imp
r
oves th
e im
a
ge b
r
ightn
e
ss to
deha
zin
g
ima
ge u
s
e
s
the
N pixel
s
in th
e field of
ste
p
brig
htne
ss
as the
brig
htness a
d
ju
stment
factor by Ret
i
nex prin
ciple
.
The experi
m
ental re
sult
s this pap
er
use
d
a lot of hazy image
s in
Yangtze
Rive
r Rip
a
ria
n
sh
ows that, can
make
m
o
st
deha
ze
d effect of image
s
more id
eal, b
u
t
can
not ap
ply all the h
a
zy
image
s,
such as im
ag
e
mutation in
h
a
zy a
r
ea. T
h
i
s
meth
od i
s
not
suitabl
e for re
al-time de
ha
zing for video
becau
se
of n
o
t runnin
g
fast enough. Ab
ove the existing
probl
em
s, in the next stage the
author team will improve effici
e
n
cy and stu
d
y
more perfe
ct
model to ada
pt to different environ
ment
hazy imag
e.
Ackn
o
w
l
e
dg
ement
The auth
o
rs
ackno
w
le
dge
the finan
cial
sup
porte
d by
the Fund
ame
n
tal Re
se
arch Fund
s
for the
Cent
ral
Unive
r
siti
es
and
Zh
eji
ang P
r
ovin
ci
al Natural S
c
ien
c
e
Fo
un
dation
of Ch
in
a
(proje
ct No.: LY12F02
01
5). The a
u
th
or is g
r
ateful
to the anon
ymous refere
e for a ca
re
ful
che
c
king of the detail
s
an
d for helpful
comment
s that improved thi
s
pap
er.
Referen
ces
[1]
Guo
w
e
i
Ya
ng,
"Preventi
o
n
of
Inlan
d
W
a
ter
T
r
a
ffic Accid
ent. Chi
n
a
W
a
ter
T
r
ansport". 9,
6 (2
009). (
i
n
Chin
ese)
[2]
Hua Z
h
ou. "Ap
p
lie
d R
e
searc
h
of Infrared Vis
i
on
D
e
tectio
n
T
e
chnolog
y"
a
nd Di
gita
l Sig
n
a
l Process
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No
. 1, Janua
ry 2013 : 362 – 3
7
0
370
in Shi
p
Nava
id
in Inla
nd. 20
08
. Chong
qi
ng U
n
iversit
y
, Ph.D.
(in Chi
nese)
[3]
RUI Yi-bi
n
; LI Peng; SU
N Jin
-
tao, "Met
ho
d of removin
g
fo
g effect from images".
Jo
urn
a
l of Co
mpute
r
Appl
icatio
ns
. 2
6
, 1 (2006). (in
Chin
ese)
[4]
F
abrizio R
u
ss
o, "An image
enh
anc
ement
techni
qu
e combini
ng sh
arpe
n
i
ng a
nd no
ise
reductio
n
".
IEEE Transactions On Instru
m
e
ntation And
Measur
em
ent
. 51, 4 (200
2)
[5]
RT
T
an. "Visibi
lit
y
i
n
ba
d
w
e
a
t
her from a sin
g
le ima
g
e
"
.
26th IEEE Confer
ence on
Com
p
uter Visio
n
and Patter
n
Re
cogn
ition
, (2
00
8) June 2
3
-2
8; Anchor
ag
e, AK, United states.
[6]
Raa
nan F
a
ttal,
"Singl
e ima
ge
deh
azin
g".
Acm T
r
ansacti
on
s On Graphics
. 27, 3 (200
8)
[7]
Kaimin
g He, Ji
an Sun a
nd
Xi
aoo
u T
ang, "Singl
e Image Ha
ze Remova
l U
s
ing Dark C
h
a
nne
l Prior".
Pattern Anal
ys
i
s
and Mach
in
e Intelli
genc
e,
IEEE Transactions on
. 33, 12 (
201
1)
[8]
LI Lo
ng-
Li; LIU
Qing; GUO J
i
an-Min
g
an
d e
t
c.,
"Defogg
ing
Alg
o
rithm of Loss
y
Compr
e
ssion
Vid
e
o
Image".
PR
＆
AI
. 24, 6 (2011
). (in Chin
ese)
[9]
Jean-P
h
il
ipp
e
T
a
rel and
Nic
o
l
as H
auti
e
re. "
F
ast visi
bi
lit
y r
e
storatio
n from
a si
ngl
e co
lor
or gra
y
lev
e
l
imag
e".
12th In
ternatio
nal C
o
n
f
erence o
n
Co
mp
uter Visi
on
, (200
9), Setp. 29 - -Oct. 2; Kyo
t
o, Japan.
[10]
Usman
Akram
.
"Reti
nal
Ima
ge Pr
eproc
es
s
i
ng: B
a
ckgr
o
u
nd
an
d N
o
ise
Segm
entati
o
n
"
.
In
do
ne
sia
Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 10, 3 (20
1
2
)
[11]
M W
o
lf and E Born. Princi
ple
s
of Optics. Electr
onics Industr
y Pu
blis
hers, Beiji
ng (2
00
5). (in Chi
nese)
[12]
Yoav Y, Sche
chner, Srin
iva
s
a G. Narasi
mhan a
nd Sh
ree K. Na
yar.
"Polariz
atio
n-
base
d
visio
n
throug
h haz
e". (200
9). Ne
w
Orlea
n
s, LA, Uni
t
ed states:
[13]
Codr
uta O. A
n
cuti, C
o
smin
Ancuti,
Chris
Herm
ans
an
d Ph
ili
pp
e Be
kaert. "A fast
semi-i
nver
s
e
appr
oach to
de
tect and remov
e
the haz
e fro
m
a singl
e ima
ge".
10th As
ia
n Confer
enc
e on Co
mput
e
r
Vision
, ACCV 201
0, Novem
b
er 8, 2010 - N
o
vember 1
2
, 20
10, (201
1) Que
ensto
w
n
, N
e
w
zeal
and
[14]
EH Lan
d an
d J
J
McCann, Li
g
h
tness a
nd reti
ne
x the
o
r
y
.
Jo
urna
l of the Optical Soci
ety of Amer
ica
. 61,
1 (197
1)
[15]
Dani
el
J. Jo
b
s
on, Z
i
a
Ur
Rahm
an
an
d
Glenn
A. W
o
ode
ll, "Pr
o
p
e
rties
and
p
e
rfor
mance
of
a
center/surro
un
d retine
x".
IEEE Transactions
On Image Pro
c
essin
g
. 6, 3 (1997)
[16]
Yuan L
U
O, Chao JI, Yi Z
hang, Z
han
gfan
g HU. "A
Sub
-
pi
xel d
e
tectio
n alg
o
rithm of
the MEMS
d
y
nam
ic fuzz
y
imag
e".
Indon
esia Jo
urna
l of Electrical E
ngi
neer
ing
. 1
0
, 8 (201
2)
Evaluation Warning : The document was created with Spire.PDF for Python.