TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 15, No. 2, August 201
5, pp. 321 ~
330
DOI: 10.115
9
1
/telkomni
ka.
v
15i2.812
5
321
Re
cei
v
ed Ma
y 5, 2015; Re
vised June
2
6
, 2015; Acce
pted Jul
y
10,
2015
Real-time Colorized Video Images Optimization Method
in Scotopic Vision
Yong Chen
*, Feng Shuai, Di Zhan
Ke
y
Lab
orator
y of Industrial In
ternet
of T
h
ings& Net
w
o
r
k Co
ntrol, MOE,
Cho
ngq
in
g Uni
v
ersit
y
of Posts
and T
e
lecom
m
unic
a
tions, C
hon
gqi
ng, Ch
in
a, 4000
65
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: chen
yo
ng@c
qupt.ed
u
.cn
A
b
st
r
a
ct
In low
li
ght e
n
v
iron
me
nt, the
survei
lla
nce v
i
de
o i
m
a
ge
ha
s low
e
r co
ntra
st
，
less infor
m
ation
an
d
unev
en bri
ghtn
e
ss. T
o
solve this pr
ob
le
m, this pa
per puts
forw
ard a contrast resoluti
o
n
compe
n
satio
n
alg
o
rith
m b
a
se
d on h
u
m
a
n
vi
sual p
e
rce
p
tio
n
mode
l. It extracts Y compo
nent fro
m
the
YUV vide
o i
m
age
acqu
ired
by ca
mer
a
ori
g
in
al
ly
to subtract co
ntrast
feature
para
m
eter
s, then
mak
e
s a pr
oporti
ona
l int
e
gral
type co
ntrast resol
u
tion
co
mpens
atio
n for l
o
w
light
pixe
ls
in Y c
o
mp
on
ent a
nd
make
s ind
e
x
contra
st
resol
u
tion co
mpens
atio
n for hig
h
lig
ht pixel
s
adaptiv
ely to
enha
nce bri
g
htness of the vide
o imag
e whil
e
ma
inta
ins the
U and V c
o
mp
one
nts. T
hen it
compresses
th
e vide
o i
m
a
ges
and trans
mits them v
i
a i
n
tern
et.
Finally,
it dec
odes and
dis
p
lays t
he v
i
deo
image on the
devic
e of in
telligent surveill
ance system
. The
exper
imenta
l
results show
that, the al
gorith
m
can effectiv
ely i
m
prov
e th
e
contrast reso
lutio
n
of the vid
e
o
imag
e a
nd
ma
intai
n
the
col
o
r of vid
eo
i
m
a
ge w
e
ll. It a
l
s
o
can
meet th
e rea
l
-ti
m
e r
e
q
u
ire
m
e
n
t of vi
de
o
mo
nitori
ng.
Ke
y
w
ords
:
vid
eo i
m
a
ge e
nha
nce
m
e
n
t, contrast resoluti
on, l
o
w
-
level-l
i
g
h
t
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
In rece
nt years, digital
image pro
c
e
ssi
ng tech
nology is widely used i
n
video
surveill
an
ce,
medical testi
ng, image
communi
catio
n
s an
d othe
r fields. The
quality of video
images captured in
appropriate illumination scen
es
will be much better.
Howev
e
r, when video
surveill
an
ce i
s
appli
ed in
actual case
s, t
he qualit
y of images is unsatisfa
ctory due to
th
e
uncontroll
abl
e environm
en
t [1, 2]. The collecte
d
im
ag
es will de
mo
nstrate d
a
rk
and low
cont
rast
overall in lo
w illumin
a
tio
n
, and hu
m
an eyes
ca
n
hardly p
e
rceive the use
f
ul informatio
n
.
Therefore,
re
coveri
ng d
e
ta
ils of low ill
u
m
inati
on ima
ge, whi
c
h h
u
m
an eye
s
ca
n hardly perceive
,
has p
r
a
c
tical
signifi
can
c
e f
o
r video surv
eillan
c
e [3].
In AINDA
NE algo
rithm p
r
opo
sed
by L
i
et al [4], g
l
obal
cont
ra
st enha
nce
m
e
n
t wa
s
impleme
n
ted
to imp
r
ove
i
m
age
qu
ality, but th
e
p
e
rf
orma
nce
can
not me
et the
re
quireme
nt for
comm
on a
ppl
ication
s
whe
n
deali
ng
wit
h
low-contrast image
s. Ta
n [5] ha
s imp
r
oved th
e im
age
quality by maximizing lo
cal cont
ra
st base
d
on a p
h
ysical deg
ra
dation mo
del
. Howeve
r, the
image tend
s
to be satu
rat
ed and a
ppe
ar halo
artifa
cts after e
n
h
anci
ng lo
cal
contrast me
rely.
The imp
r
ove
d
histog
ram
equali
z
ation
algorith
m
pr
o
posed by Li
CH et al [6] is prefe
r
a
b
ly
maintaine
d
the col
o
r info
rmation of the
image to so
me extent, but the enha
n
c
eme
n
t effect is
better fo
r
a
relatively narrow
ra
nge
of
gray
scale. I
n
[7], a vi
su
al comp
en
sat
i
on m
e
thod
wa
s
pre
s
ente
d
b
a
s
ed
on th
e
model
NR-IQ
A
, combi
n
ing
with the
re
solution
cha
r
a
c
teri
stics of t
he
human
eyes
to comp
en
sat
e
the imag
e informatio
n, this alg
o
rithm
sho
w
s goo
d
perfo
rman
ce
in
low lig
ht ima
ge sce
n
e
s
, b
u
t not for
su
ch scen
e which co
ntain
s
lo
w-lig
ht and
hi
gh-lig
ht re
gio
n
or
has un
even l
u
minan
ce.
Literatu
re [8]
de
veloped
a
Re
tinex image
e
nhan
cem
ent
algorith
m
b
a
sed
on Zerni
k
e m
a
trix, which h
a
s mo
re se
nsitivity in
enhancin
g the dynamic imag
e a
t
a wider ra
ng
e,
but it is inte
nsive
com
put
ationally an
d
difficult
to
meet the
re
q
u
irem
ents of
real
-time vi
deo
pro
c
e
ssi
ng. An enha
nceme
n
t algorithm
wa
s propo
se
d based o
n
the dark chan
n
e
l prio
ri low li
ght
image [9]. T
h
is m
e
thod
i
m
prove
s
th
e
enha
ncement
effect a
nd
redu
ce
s the
complexity of
the
algorith
m
, but it is still hard
to adapt
to the real
-time video processi
ng.
Based
on
th
e analy
s
e
s
mentione
d a
bove, a
cco
rd
ing to the
chara
c
te
risti
c
s of video
surveill
an
ce i
n
cha
nging lo
w-lig
ht enviro
n
ments,
the
method which combi
ned
color ima
ge qu
ality
evaluation m
odel a
nd ima
ge en
han
cem
ent algo
rithm
s
was
pro
p
o
s
ed to optimi
z
e ea
ch d
a
rk
and
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 321 –
330
322
bright visi
on
pixel based o
n
the hum
an
visual
cha
r
a
c
teristics, an
d
usin
g self
-opt
imizing
co
ntrast
resolution
parameter
com
p
ensation met
hod can av
oi
d the tediou
s
manual
settin
g
s. Experim
e
n
ts
sho
w
that thi
s
meth
od
ca
n
effectively e
nhan
ce th
e vi
deo im
age
s
capture
d
in
lo
w light
scen
e
and
keep the ori
g
inal image
color information more
to improve the capab
ility of real-tim
e video
pro
c
e
ssi
ng.
Well, the
rest
of this pa
pe
r i
s
o
r
ga
nize
d a
s
follo
ws. In Se
ct.2, we revie
w
th
e
base of
image optim
izing a
nd e
v
aluation wit
h
Sect.3
describi
ng ad
aptive contrast re
solutio
n
comp
en
satio
n
meth
od. S
e
ct.4 give
s the h
a
rdwa
re
syste
m
, an
d
Sect.5
provides a
set
of
test
results. The
concl
u
si
on follows up after t
h
is sectio
n.
2. Basic Image Optimiza
tion and Ev
aluation
2.1. Contr
a
s
t
Resolu
tion Cons
train
t
s
The literatu
r
e [10, 11] studied
contrast re
solutio
n
with the variation of the gray
backg
rou
nd.
For [0,25
5
] g
r
ayscale im
a
ge, m
ean
s th
at gray value
is between
0 and 4
7
, an
d
photoni
c is b
e
twee
n 48 an
d 255. For dif
f
erent opti
c
al
area, the mini
mum gradati
on differen
c
e
of
human
visio
n
ch
ang
e al
on
g with
gray b
a
ckgroun
d,
b
u
t it is
non
-lin
ear. F
o
r the i
m
age i
n
scotopic
regio
n
, part o
f
the target image inform
ation whi
c
h i
s
maske
d
by low-contrast
backgroun
d, can
not be
ide
n
tified by th
e
hu
man
eyes.
Th
erefo
r
e, in
thi
s
p
ape
r, the
i
m
age
ta
rg
et c
h
rom
a
ticity
le
v
e
l
whi
c
h i
s
dif
f
i
cult
t
o
di
st
in
guis
h
w
a
s j
u
st
co
mp
e
n
sa
ted to the ex
tent of the h
u
man
eyes
can
disting
u
ish, so as to achie
v
e mining co
nce
a
l inform
a
t
ion and imp
r
oving image
quality.
2.2. Zadeh T
r
ansform
The mi
nimum
differe
nce of
gray in
an
i
m
age
that h
u
m
an vi
sual
contra
st resolu
tion can
disting
u
ish called JND
(Ju
s
t
Notice
a
b
le Differen
c
e)
[7]. Hum
an
visual co
ntrast
re
sol
u
tion
nonlin
ear
co
mpen
sation
p
r
inci
ple i
s
tha
t
expandin
g
a
gray level
be
tween
adja
c
e
n
t pixels to
o
n
e
JND, then u
s
eful informati
on is
ma
sked
by the backgrou
nd so that
they can b
e
resolved to
the
extent that can be distin
gui
she
d
.
2.3. Color Image Quality
Ev
aluation Index
Huma
n visi
on
pro
b
lem
s
a
r
e psy
c
hol
ogi
cal physi
cal
problem
s; the i
m
age
quality
is al
so
a psychologi
cal evaluatio
n of physica
l problem
s [12]. Quoting
comp
rehe
nsive color im
ag
e
quality evalua
tion function
expre
ssi
on (1
) from [13, 14
].
ABWF
×
AHF
×
NNF
×
APCL
×
AIE
=
CAF
(1)
Whe
r
e AIE, APCL, AHF,
NNF
and A
B
WF re
present the information entro
py, the
physi
cal
cont
rast, ave
r
a
g
e
level fa
cto
r
, the av
e
r
a
ge b
r
ightn
e
ss n
o
rm
alizati
on a
r
e
clo
s
e
to
distan
ce and averag
e
ba
n
d
width
fa
ctor respe
c
tive
ly. The big
g
e
r
th
e CAF valu
e i
s
, the bette
r the
image qu
ality of human visual perce
ptio
n will be.
3. Adap
tiv
e
Con
t
ras
t
Re
solution Co
mpensa
tion Metho
d
3.1. Video Image Co
ntr
ast Resolu
tion Compen
sati
on
The format of
video imag
e
colle
cted
by the front
-en
d
came
ra i
s
u
s
ually YUV, where Y i
s
the lumin
a
n
c
e or
bri
ghtne
ss, a
nd
U a
n
d
V are
the
R-Y a
nd B
-
Y com
pon
ent resp
ectively, also
kno
w
n as ch
r
o
ma
w
h
ic
h
d
e
s
c
rib
e
s col
o
r
s
a
t
u
rat
i
o
n
a
ttribute [15]. The a
d
vanta
ge of YUV i
s
that
luminan
ce
si
gnal (Y
) an
d ch
romin
a
n
c
e
signal
(U, V) are mutual ind
e
pend
ent and
two
comp
one
nts U
a
n
d
V
ca
n
represent colo
r whi
c
h
mean
s com
p
res
s
io
n,
t
r
a
n
smis
sio
n
and
pro
c
e
ssi
ng
a
r
e m
o
re
ea
si
er. Ma
kin
g
u
s
e
of tho
s
e
YUV features, a compe
n
sation meth
od
is
prop
osed to
optimize Y
co
mpone
nt with
out losi
ng
col
o
r info
rmation
of image
s while the U
an
d V
are con
s
tants.
Based
on
ch
ara
c
teri
stics
of hum
an vi
sion
cont
rast
resol
u
tion limit
, it com
b
ine
s
the JND
with pro
portio
nal integral operatio
n of automatic
cl
osed-lo
op control system
to optimize the
pixel
who
s
e
gray scale
value
ra
nge from 0
to
47
(in
sc
oto
p
ic
regi
on) u
s
ing p
r
op
ortio
nal-inte
g
ral-type
contrast resol
u
tion com
pen
sation o
p
timization method.
The equatio
n
is sho
w
n a
s
Equation (2).
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Real
-tim
e Col
o
rized Vide
o Im
ages Optim
i
zation Meth
o
d
in Scotopi
c Vision (Yo
ng
Che
n
)
323
)
,
,
(
0
)
(
)
,
,
(
Y
y
x
OG
i
i
i
JND
k
Y
y
x
T
(2)
Whe
r
e T
(x, y, Y) represent
s the targ
et lu
minan
ce valu
e of the com
p
ensated da
rk vision
pixels in Y
co
mpone
nt ima
ge, OG
(x, y,
Y) rep
r
e
s
e
n
ts the origi
nal l
u
minan
ce
of scotopi
c pixel
in
Y compo
nent
image. k re
p
r
esents the
compen
satio
n
coeffici
ent wh
ich is a p
o
siti
ve real num
b
e
r,
and JND (i) repre
s
e
n
ts Ju
st Noticeabl
e brightn
e
ss di
stan
ce on the
backgroun
d brightn
e
ss i.
If only comp
ensate Y
co
mpone
nt of
the da
rk visi
on pixel,
im
age l
a
yerin
g
will
be
redu
ce
d, ima
ge inform
atio
n will be imb
a
lan
c
e
an
d the image
will
be over en
h
anced pa
rtially.
Therefore, th
e Y comp
one
nt of photopi
c vision
also
need
s to be
optimize
d
an
d com
pen
sat
ed.
Ju
st noticea
bl
e resolution o
f
clear visual i
s
in
rang
e at 1.17-1.7
5
[6]. Based on hu
man visual h
a
s
a certai
n a
d
a
p
tive ch
aract
e
risti
c
s to vi
sual im
age
s,
combine
d
th
e
global
map
p
i
ng m
e
thod,
more
in line
with
th
e hu
man
eye
for
high
dyn
a
mic imag
e i
dentificatio
n,
with exa
c
tly resolva
b
le
JND.
Acco
rdi
ng to
Equation
(2
), esta
blish ex
pone
ntial
con
t
rast
re
solutio
n
compe
n
sation o
p
timizati
on
method un
de
r clea
r visu
al as Equatio
n (3) sh
ows.
r
Y
y
x
OG
i
i
JND
Y
y
x
OG
Y
y
x
T
)
)
i
(
(
)
,
,
(
)
,
,
(
)
,
,
(
0
(3)
Whe
r
e T
(x, y, Y) rep
r
e
s
ent
s the ta
rget l
u
mi
nan
ce val
ue of the
co
mpen
sated
d
a
rkvi
sion
pixels of th
e
Y comp
one
nt image, O
G
(x, y, Y)
rep
r
ese
n
ts the
original lumi
na
nce
of scoto
p
ic
pixel of th
e
Y com
pon
ent
imag
e, r re
pre
s
ent
s
a v
a
riabl
e p
a
ra
meter
adju
s
ti
ng
com
pen
satio
n
depth,
a
nd JND (i) re
prese
n
ts exact
l
y
resolv
able
brig
htne
ss
distan
ce
on
the ba
ckgrou
nd
brightn
e
ss i.
For the im
ag
e, the avera
ge bri
ght
ne
ss is i
n
da
rk
vision a
r
ea,
Whe
n
the pi
xel
luminan
ce va
lue of the Y compon
ent image less than
or equal 47,
we u
s
e Equa
tion (2) to ca
rry
out propo
rtio
nal-inte
g
ral-type cont
rast
resol
u
tion
co
mpen
sation
u
nder the d
a
rk vision.
Whe
n
the
pixel brig
htne
ss val
u
e
s
of the Y comp
on
ent imag
e
greater th
an 4
7
and l
e
ss tha
n
equ
al to 2
5
5
,
we
use Equ
a
tion (3) to
carry o
u
t pro
portion
al-inte
g
ral
-
type
con
t
rast
re
soluti
on
comp
en
sation
unde
r the cl
e
a
r vision. Accordin
g to the above,
co
ntra
st re
solution
comp
en
satio
n
wa
s propo
sed
for colo
r imag
e, CRCCI a
s
Equation (4)
sho
w
s.
(4)
To avoid the
phen
omen
on
of anti-col
o
r
compl
e
me
nta
r
y colo
r that origin
al bri
ght
er pla
c
e
will be dimm
ed and defo
r
mity after compen
sation o
p
timized. Fo
r the pixels that compe
n
sated
target
brig
htn
e
ss valu
es
greater than
25
5, set
th
e val
ue to
255,
an
d If the valu
e
is l
e
ss th
an
0,
set the value
to 0. Equation
(5)
sho
w
s the co
nstraint
s:
(5)
Due to the u
n
co
ntrolla
ble
nature of lighti
ng con
d
itio
ns in video
surveill
an
ce
and the
contin
uity of the video [16
], to make e
a
ch f
r
ame to
achi
eve opti
m
um ad
aptive com
pen
sati
on
effect, sele
ct
the imag
e m
ean va
rian
ce
whi
c
h
can
b
e
better
ch
aracteri
ze
imag
e pa
ramete
rs as
the ada
ptive para
m
eter. E
quation
(6
) is a mea
n
sq
u
a
re
error ch
a
r
acte
ri
zing th
e imag
e feat
ure,
Whe
r
e K (i, j) is the image
averag
e gray.
(,
,
)
0
(,
,
)
0
(,
,
)
(,
,
)
0
(,
,
)
(
)
(
,
,
)
4
7
(
,
,
)
(
)
47
(
,
,
)
255
iO
G
x
y
Y
i
r
iO
G
x
y
Y
i
OG
x
y
Y
O
G
x
y
Y
Tx
y
Y
k
J
N
D
i
O
G
x
y
Y
OG
x
y
Y
J
N
D
i
O
G
x
y
Y
0,
(
,
,
)
0
(,
,
)
255
,
(
,
,
)
255
Tx
y
Y
Tx
y
Y
Tx
y
Y
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 321 –
330
324
2
1
0
1
0
||
)
,
(
)
,
(
||
1
ar
N
j
M
I
j
i
K
j
i
I
N
M
V
(6)
3.2. Adap
tiv
e
Con
t
ras
t
Resolution Pa
ramete
r Selection
Thro
ugh exp
e
rime
nts respectively buil
d
t
he rel
a
tio
n
shi
p
s b
e
twe
en k
(compe
nsatio
n
coeffici
ent) a
nd mea
n
squ
a
re e
r
ror
u
n
d
e
r da
rk vision
, the relation
ship bet
ween
R (com
pen
sa
tion
coeffici
ent) a
nd the me
a
n
sq
uare un
der
clea
r vi
sion.
Sele
ct a pictu
r
e
with a lo
w-li
ght
environ
ment
as expe
ri
mental m
a
terial
f
r
om
the
Colo
r Ch
ecke
r
Data
set G
a
llery
(http://www.eecs
.
harvard.edu/~
a
ya
nc
/oldcc
/dbs
.html). By c
ons
tantly dis
t
urbing the
comp
en
satio
n
co
efficient
k an
d r to lo
ok fo
r the
m
a
ximum valu
e of the
obje
c
tive pa
rame
ters
(CAF
) rep
r
e
s
enting the im
age quality, and com
b
ine
d
with su
bjectiv
e
evaluation to determin
e
the
best
quality i
m
age
after
o
p
timization. A
s
sho
w
n
by
t
he
cu
rve fitting, Figu
re1
an
d Fig
u
re
2
sh
ow
this
relationship.
Figure 1. Rel
a
tionship bet
wee
n
k an
d mean
squ
a
re d
e
viation in scotopi
c vision
Figure 2. Rel
a
tionship bet
wee
n
r and m
ean
squ
a
re a
nd m
ean squa
re e
rro
r in photo
n
i
c vision
Thereby e
s
t
ablishing
a
function,
as E
quatio
n (7)
sh
ows, b
e
twee
n
com
pen
satio
n
coeffici
ent k a
nd the mean
squ
a
re e
r
ror
of t
he original
image und
er
dark vision
co
ndition
s.
113
.
0
/
138
.
5
Var
k
(7)
The sam
e
m
e
thod can
b
e
u
s
ed
to perfo
rmed r
i
ndex compen
satio
n
coeffici
ent
a
u
t
omatic
optimizatio
n
on the o
r
igin
al image u
n
d
e
r cl
ear vi
sio
n
con
d
ition
s
. Cre
a
te coefficient r
and t
he
mean squa
re
error of the original
imag
e as Equatio
n (8) sh
ows.
0308
.
0
0117
.
0
10
4645
.
2
r
2
4
Var
Var
(8)
3.3. Adap
tiv
e
Con
t
ras
t
Resolution
Co
mpensa
tion of the Video
Image
First, calcul
ate mean
squ
a
re error fo
r ea
ch fram
e of video ima
ge ca
ptured from the front
came
ra. The
n
, calculate k, r value of each fram
e im
age by the m
ean squa
re e
rro
r; optimize
the
Y com
pon
en
t of ea
ch frame im
age
by adaptive
comp
en
satio
n
optimi
z
atio
n. Hol
d
ing
U, V
comp
one
nts
of the video
stream a
r
e fo
r
the sa
me
to
give a final e
nhan
ce
d vide
o image
of YUV
spa
c
e. Fin
a
ll
y, convert the enha
nced
video image i
n
to an RGB colo
r imag
e and sho
w
it in the
monitori
ng eq
uipment.
4. Hard
w
a
re
s
y
stem
As
F
i
gu
re
3
sh
o
w
s
,
th
e
s
yste
m
co
mpr
i
ses
a pr
oc
es
s
i
ng
(D
M64
6
7
)
un
it a
n
d
a vid
e
o
in
pu
t
units.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Real
-tim
e Col
o
rized Vide
o Im
ages Optim
i
zation Meth
o
d
in Scotopi
c Vision (Yo
ng
Che
n
)
325
Figure 3. Hardwa
re sy
ste
m
diagram
4.1. The Experimental Re
sults and Su
bjectiv
e Ev
aluation
The TM32
0DM6467
T wa
s used a
s
main pro
c
e
s
so
r chip of the
core of the
system
pro
c
e
ssi
ng u
n
it,
the
chip integrate
s
a high-
pe
rform
ance TMS
3
2
0
C6
4x + DS
P co
re
and
an
ARM92
6
EJ-S co
re. And th
e clo
c
k fre
q
u
ency
rea
c
he
s 1GHz, in
whi
c
h
DSP is m
a
inly respon
sib
l
e
for co
mpression and
rel
a
ted video d
a
ta algo
rithm
,
and ARM
is pri
m
arily resp
on
sible f
o
r
comm
uni
cati
on with pe
rip
heral
s an
d rel
a
ted sche
duli
ng.
4.2. Video Input and O
u
tput Uni
t
Video captu
r
e
mod
u
le co
n
s
ist
s
of a CCD came
ra
an
d a Xilinx Vi
rtex-4 FP
GA p
r
ocesso
r.
The video st
ream
captu
r
ed by acqui
sition mod
u
l
e
is pa
ssed
to the FPGA processo
r b
y
TVP5150,
an
d FPGA
will
compl
e
te th
e
switchi
ng
a
m
ong
interfa
c
e
s
Y
C
b
C
r /
HDMI
/ HDS
D
I /
VGA. To improve the colle
ction efficie
n
cy, use ti
me-multiplexing t
o
ensure mul
t
i-cha
nnel
sig
nal
input. Tran
sp
ort data to DDR2 for the u
s
e of DM
646
7T throu
gh DMA mode.
5. Integra
t
ed
Experiment
Software de
sign of the system is bas
ed on an e
m
bedd
ed Lin
u
x operatin
g
system
appli
c
ation
software
platfo
rm, incl
udin
g
video c
aptu
r
e, video p
r
e-pro
c
e
ssi
ng, video p
r
o
c
e
ssi
ng
algorith
m
s a
n
d
comp
re
ssio
n, network tra
n
smi
ssi
on, the client de
co
des di
spl
a
y module.
The expe
rim
ental syste
m
captu
r
e
s
video
st
ream
with CCD
ca
mera e
quip
m
ent and
follows Vide
o
4Linux2
(refe
rre
d V4
L2) frame flo
w
, ta
ke
s o
u
t the
origin
al vide
o
stream
in th
e
frame buffer t
h
rou
gh get
Ca
ptureBuffer(),
calculat
es th
e overall average varia
n
ce for each fram
e
image, the k, r values a
r
e o
b
tained by ca
lculatin
g the averag
e varia
n
ce of the fra
m
e image, an
d
extract
s
the
Y comp
one
nt to make
ada
ptive co
mp
en
sation, mai
n
tains th
e U, V
comp
one
nts of
the video
stream fo
r the
same, th
en t
he e
nhan
ce
d
image
is obt
ained i
n
the
YUV spa
c
e.
Call
cod
e
c of
DSP side by Cod
e
c En
gin
e
mec
hani
sm, call H.26
4 encode
r for en
co
ding
the
pro
c
e
s
sed vi
deo st
ream,
then the dat
a wa
s tran
smitted throug
h RTP net
work
proto
c
ol
and
displ
a
yed after de
codi
ng i
n
monitori
ng termin
al.
Specific ste
p
s
: First, initi
a
lize
the
en
gi
ne th
rou
g
h
calli
ng th
e
initialization
modul
e
CERu
ntime_i
nit () of the
core
engi
ne A
P
I; Then o
p
e
n
the e
n
codin
g
en
gine
and
return a
hEng
ine
pointer th
rou
gh calli
ng En
gine_
open
();
Followed,
create a co
ded
handle
ca
ll
s by Venc1
_
create
() th
rou
gh
cal
ling hEngi
ne
pointer; At la
st thro
ugh
cal
ling Ven
c
1_
p
r
ocess
() to
a
c
hieve th
e video
strea
m
en
co
ding by H.2
6
4
en
code
r, then the e
n
coded vide
o stream i
s
tran
smitted ove
r
the
netwo
rk an
d t
hen
displaye
d in th
e
client
after
de
cod
e
r
d
e
coding.
Experime
n
tal d
e
sig
n
flo
w
ch
art
is sh
own in Figure 4.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 321 –
330
326
Figure 4. Experim
ent de
si
gn flow chart
5.2. Analy
s
is
of Experime
ntal Re
sults
The imp
r
ove
d
bidirectio
na
l histog
ram
(abbrevia
ted:
I-HE) [17], an
improve
d
m
u
lti-scale
Retinex (abb
reviation: I-MSR) [18
-
2
0
] an
d co
nt
ra
st re
solution comp
ensation met
hod (refe
rre
d
to
as:
CRC) we
re u
s
e
d
to
co
mpare. The
experim
ent
al
environ
ment i
n
clu
d
e
s
three
kin
d
s
of vide
o
images (video surveill
ance sites:
http://www.vcn.
c
om.cn/) and a set of video with different
light,
different traffic
monitoring
s
i
tes
.
1) Subjectiv
e
e
v
a
l
uation
Experiment o
ne: Figu
re 5
(
a) sho
w
s
a
video
und
er l
o
w-li
ghting traffic co
nditio
n
s. The
chroma
spe
c
t
r
a an
alysi
s
sh
ows that this
video fram
e
chrom
a
level i
s
rel
a
tively co
nce
n
trated, lo
w
contrast,
narrow
ban
dwi
d
th, useful info
rmation
is
dif
f
icult to
be i
d
entified by th
e hu
man
eye
s
.
Figure 5(b)
sho
w
s an i
m
age
whi
c
h
wa
s proces
sed
by improved bidi
re
ctional hi
stog
ram
equali
z
ation.
The entire im
age chroma l
e
vel has
bee
n stret
c
he
d, but the tensil
e ran
ge of hi
gh
den
sity chro
minan
ce leve
l is large
r
tha
n
the ra
nge
of low-den
sit
y
chro
mina
nce level and t
here
has be
en th
e
image
contrast
stret
c
hing
uneve
nne
ss, re
sulting
in
colo
r a
bno
rm
al an
d
severe
colo
r di
stortio
n
. Figure 5(c) shows the
i
m
age
p
r
o
c
e
s
sed by
multi-scale Retine
x
enhan
cem
ent
method, a
c
co
rding
to the
o
p
timization
ef
fect, ch
rom
a
l
e
vel imag
e h
a
s
bee
n effe
ctively stretche
d,
and
details o
f
the imag
e a
r
e hi
ghlig
hted
. But a lot of
chroma
spe
c
tral a
r
e
betwe
en 1
00
- 1
6
0
,
whi
c
h m
a
ke
the ove
r
all
image
too
bright
and
a
ppea
rs whitish
phen
ome
non, the ve
hicle
operating con
d
itions is diffi
cult to identify in
the figure. Figure 5
(
d)
shows the i
m
age which wa
s
pro
c
e
s
sed by
cont
rast
re
solution
comp
ensation,
ima
ge layeri
ng p
r
ocesse
d by
this metho
d
i
s
signifi
cantly stronge
r, ch
ro
ma spe
c
tral
distrib
u
tion is more uniform, and the color of the image
has be
en we
ll maintained,
the color p
r
oportio
n
of the vehicle an
d the road surface be
co
mes
coordination to facilitate real-time monit
o
ring.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Real
-tim
e Col
o
rized Vide
o Im
ages Optim
i
zation Meth
o
d
in Scotopi
c Vision (Yo
ng
Che
n
)
327
(a) O
r
igin
al image
(b) Imp
r
oved
two-di
men
s
io
nal histo
g
ra
m
(c) Improve
d
Multi-scal
e Retinex
(d)
Contrast
Re
solutio
n
Compen
satio
n
Figure 5. Traf
fic re
cord and
2 level chro
ma spe
c
tra
Experiment t
w
o: Fo
r u
nev
en illumi
natio
n app
ea
red i
n
video
su
rveillan
c
e, such a
s
the
ca
se of point
light sou
r
ce i
n
Figure 6, a partial
regio
n
of the image
appe
ar
s in strong light, and
a
partial regio
n
of the image ap
pea
rs in lo
w ligh
t. Process the image th
roug
h impro
v
ed
bidire
ction
a
l histog
ram, al
though
the b
r
ightne
ss
of the e
n
tire
ima
ge
wa
s e
nha
nce
d
, ho
wev
e
r,
the lighter are
a
s of th
e
processed im
age
occurre
d
the
effo
rt of over
enha
ncement
, and the lo
wer
illumination region enh
an
ced obviou
s
l
y
,
a
colo
r
d
i
stortion
o
c
curs. The
im
age whi
c
h wa
s
improve
d
by
Multi-scal
e Retinex
processing
wa
s
sh
o
w
n in
the Fi
g
u
re
6(c). T
h
e
overall
imag
e
occurre
d
ove
r
en
han
ce
d, the pictu
r
e
a
ppea
red
wh
itish a
nd th
e stations
wa
s
more
difficult
to
identify. Figu
re 6
(
d
)
sho
w
s the im
ag
e after
co
ntrast resolutio
n
compe
n
sation p
r
o
c
e
ssi
ng,
chroma
sp
ect
r
al di
strib
u
tio
n
is mo
re
uni
form, maintai
n
s g
ood
col
o
r, and le
ss
det
ail is lo
st, mo
re
in line with th
e human eye.
Experiment three: the ori
g
inal image in Fi
gure 7-8 was with un
even illuminatio
n and the
lowe
r overall
image b
r
igh
t
ness, after
optimize
d
by
the improve
d
bidirectio
n
a
l histog
ram
the
image
sh
own
in Fig
u
re 7
(
b
)
, 8(b) app
ea
red u
neven
di
stributio
n of li
ght an
d d
a
rk,
pavement
col
o
r
imbalan
ce
ca
use
d
p
oor o
v
erall visual
effect
of the i
m
age. Th
e i
m
age
whi
c
h
wa
s p
r
o
c
e
s
sed by
improve
d
Mul
t
i-scale
Retin
e
x pro
c
e
ssi
n
g
wa
s
sho
w
n
in Figu
re 7
(
c) an
d 8(c), the overall
ima
ge
occurre
d
over enhan
cem
e
n
t, more imag
e information
is lost. The image
s whi
c
h
were p
r
o
c
e
s
sed
by
contrast resol
u
tion co
mpen
sation shown
in
Fig
u
re 7(d
)
a
nd 8
(
d), imag
e inf
o
rmatio
n is g
ood,
informatio
n i
s
evenly di
stributed, colo
r inform
ation
well mai
n
ta
ined, an
d p
a
vement col
o
r
uniformity is
more
suitabl
e
for the huma
n
eye to obse
r
ve.
(a) O
r
igin
al image
(b) Imp
r
oved
two-
d
i
me
ns
io
na
l
histog
ram
(c
) Improv
e
d
Multiscale Retinex
(d) Contr
a
st
Re
solutio
n
Comp
en
satio
n
Figure 6. Gas station effect
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 321 –
330
328
(a) O
r
igin
al image
(b) Imp
r
oved
two-
d
i
me
ns
io
na
l
histog
ram
(c
) Improv
e
d
Multiscale Retinex
(d) Contr
a
st
Re
solutio
n
Comp
en
satio
n
Figure 7. Hig
h
spe
ed car e
ffect
2) Objectiv
e
e
v
aluation
Select info
rm
ation ent
ropy
, contrast, a
nd CAF
obje
c
tive indi
cato
rs to
cal
c
ul
ate thre
e
grou
ps of video image an
d a set of video, after
proce
s
sed by two-way
hi
sto
g
ram, improved
multi-scal
e Retinex
an
d contra
st
resol
u
tion co
mp
e
n
satio
n
meth
ods.
The
val
ues are
sho
w
n i
n
Table 1, Tabl
e 2. After using cont
ra
st reso
l
u
tion co
mpen
sation, i
m
age inform
ation entro
py and
contrast a
s
well as the
CAF value are
significa
nt
ly larger tha
n
othe
r method
s, which
sho
w
s the
con
s
i
s
ten
c
y of subje
c
tive
and obje
c
tive evaluation, and indicates that the video whi
c
h
pro
c
e
s
sed by
the propo
se
d
treatment m
e
thod
s ha
s b
e
tter quality.
Table 1. Vide
o image qu
ali
t
y asse
ssmen
t
Method
aramete
r
Information Ent
r
o
p
y
Contrast
CAF
Fig.5
I-HE
5.136
1.145
5.125
I-MSR
5.867
0.766
4.367
CRC
7.577
2.541
19.146
Fig.6
I-HE
6.956
2.339
8.584
I-MSR
6.835
1.587
5.043
CRC
7.322
2.403
15.099
Fig.7
I-HE
6.088
2.231
5.400
I-MSR
6.362
1.970
6.569
CRC
6.449
2.971
10.324
3) Real-time
v
e
rification
To validate the real time
of the algorit
hm,
Figure 8
is the 100 frames of vide
o traffic
monitori
ng.
With respe
c
t to the imp
r
ov
ed bidi
re
ction
a
l histo
g
ram
equali
z
ation
(Figure 8
(
b)),
the
effect of the image p
r
o
c
e
s
sed by the i
m
prove
d
mu
l
t
i-scale
Retin
e
x algorithm
(Figu
r
e 8(c)) is
more
si
gnificant. Ho
weve
r, this m
e
thod
use
s
li
nea
r it
erative m
ann
er to
pe
rform
the time-dom
ain
filtering for th
e illumination
comp
onent
of the im
age
which ha
s large a
m
ou
nt of computati
on.
The average time of
processi
ng
a
surveillance
video
with the i
m
proved mul
t
i-scale
Retinex
algorith
m
is 0
.
18s, and the
frame rate is about 5
fram
es/sec. Bidire
ctional hi
stog
ram processi
ng
has faste
r
sp
eed, b
u
t it al
so
nee
ds to
cou
n
t vi
deo
f
r
ame
pixel
probability di
stribution
and
d
o
it
twice, so
the
averag
e
time
of
processin
g
ea
ch
fram
e t
a
ke
s
92m
s, a
frame
rate
of 10
fram
es/
s
e
c
.
Contrast
resolution comp
ensation, du
e to the
com
pen
sation o
p
t
imization me
thod is
relati
vely
simple, the
time complexit
y
of the algo
ri
thm is lo
w, th
e avera
ge tim
e
of
processi
ng ea
ch f
r
am
e is
48ms, f
r
ame
rate i
s
2
0
fra
m
es/
s
e
c
, whi
c
h i
s
si
g
n
ificantly better t
han th
e oth
e
r two
method
s in
orde
r to meet
the requi
rem
ents of re
al-ti
m
e in
video p
r
ocessin
g
as
sho
w
n in Ta
b
l
e 2.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Real
-tim
e Col
o
rized Vide
o Im
ages Optim
i
zation Meth
o
d
in Scotopi
c Vision (Yo
ng
Che
n
)
329
Table 2. Re
al
-time video i
m
age evalu
a
tion paramete
r
s
Method Param
e
t
e
r
Information
Entrop
y
Contras
t
CAF
Frame
rate (f
ram
e
s
/ s
e
c
)
Fi
g 8
I-HE
5.245
2.396
1.879
10
I-MSR
7.075
2.358
11.145
5 frames / sec
CRC
7.182
2.438
15.800
20
frames / sec
6. Conclusio
n
Based
on
th
e hum
an vi
sual
cha
r
a
c
teristic, u
nde
r t
he lo
w ill
umi
nation, p
r
o
c
e
s
s the
pixels i
n
diff
erent
area
s
of the vide
o
image
thro
ugh
different
types
of co
ntrast
re
sol
u
tion
comp
en
satio
n
. And throu
gh self-optimi
z
ing o
p
ti
mal comp
en
satio
n
para
m
eters, maintain U,
V
comp
one
nt unch
ang
ed, so
that
the brightness of the image has b
een enh
an
ce
d to ensu
r
e the
image
colo
r
saturation. Fi
nally by codi
ng tran
smi
s
si
on net
work,
the image i
s
decode
d an
d
displ
a
yed i
n
the
clie
nt a
nd fo
rm
a
complete
mon
i
toring
sy
ste
m
. Throug
h
the an
alysi
s
of
subj
ective
an
d obj
ective
e
v
aluation, Th
e propo
se
d a
l
gorithm
in th
e overall e
n
h
ancement
of
the
video ima
ge,
colo
r m
a
intai
n
ing
and li
ght
re
cove
ry
hav
e a
c
hieved
g
ood
re
sults. And
the average
time of this a
l
gorithm
pro
c
essing vid
eo
is 48m
s,
whi
c
h compli
es with
the requi
reme
nts
of re
al-
time video su
rveillan
c
e to ensure the ov
erall effect of
the video.
Ackn
o
w
l
e
dg
ements
Authors
woul
d like to th
an
k the
Chon
gq
ing
Edu
c
atio
n
Co
mmittee Scien
c
e of China
fo
r
sup
portin
g
the Found
ation
of prog
ram, No KJ140
043
4
.
Referen
ces
[1]
CHEN
D, T
I
AN
F
C, LIU Y, H
U
Y W
,
HAN
L. Im
age E
n
h
anc
ement i
n
C
o
h
e
r
ent Optica
l A
m
plificati
o
n
b
y
Photorefractiv
e
Cr
y
s
tals.
Jo
urnal of Co
mpute
r
and Co
mmun
i
catio
n
s.
201
4; 2: 42-47.
[2]
NAYAK DR,
BHOI A. Ima
ge En
hanc
em
ent Usin
g Fu
zz
y
Mor
pho
lo
g
y
.
Jo
urn
a
l o
f
Engin
eeri
n
g
Co
mp
uters & Appli
ed Sci
enc
e
s
.
2014; 3(3): 2
2
-26.
[3]
Benn
ett EP, McMilla
n L. Vid
e
o
enh
anc
eme
n
t
using p
e
r-pi
x
el virtua
l e
x
p
o
s
ures.
ACM T
r
ansiti
ons o
n
Graphics(T
OG).
2005; 24(
3): 845-
852.
[4]
LI
T
,
ASARI V.
An integrate
d
nei
ghb
orh
ood
dep
en
dent ap
p
r
oac
h for no
nli
near e
nha
nce
m
e
n
t of colo
r
imag
es
. Proc
eedings
of the IEEE Co
m
puter S
o
ciet
y
Internationa
l Conferenc
e
on Inform
atio
n
T
e
chnolog
y: C
odi
ng a
nd Com
putin
g-IT
CC. 2
004; 2: 13
8-13
9.
[5]
T
A
N K, OAKLEY PJ. Ph
ysics-bas
ed
ap
proac
h to c
o
l
o
r ima
ge
en
h
ancem
ent i
n
poor v
i
sib
ilit
y
cond
itions.
Opt
i
cal Soc
i
ety of Amer
ica
. 20
01;
18(10): 24
60-
246
7.
[6]
LI CH, LIU H
,
Z
H
ONG C.
Appl
icatio
n of
histogram m
odific
a
tion
in i
m
age e
n
h
anci
ng.
J. Hube
i
Univers
i
ty of Techn
o
.
201
1; 26(2): 67-7
0
.
[7]
Stark JA. Ada
p
tive im
ag
e c
ontrast e
nha
n
c
ement
usin
g gen
eral
izati
ons
of
histo
g
ram
equ
aliz
atio
n.
IEEE Trans. Image Proc
essin
g
.
2000; 9(
5): 889-8
96.
[8]
CHEN Y, CHE
NG JJ, XIE ZHX. A comp
ens
a
t
ion
meth
od of
huma
n
visu
al s
y
stem b
a
se
d –
on NR-IQA.
Journ
a
l of Ch
o
ngq
ing U
n
iv
ers
i
ty
. 2013; 36(
2)
: 141-14
7.
[9]
W
A
NG RG, ZHANG XT
, Z
H
ANG X, F
A
N
G
S. A novel
Retin
e
x
al
gorit
hm for ima
ge
enh
anc
ement
base
d
on Z
e
rn
i
k
e moment.
Jo
urna
l of Imag
e
and Grap
hics.
201
1; 3(16): 31
0-31
5.
(a) O
r
igin
al image
(b)Imp
rov
e
d two-
d
i
me
ns
io
na
l
histog
ram
(c
) Improv
e
d
Multiscale Retinex
(d) Contr
a
st
Re
solutio
n
Comp
en
satio
n
Figure 8. Traf
fic light intersection effe
ct
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 15, No. 2, August 2015 : 321 –
330
330
[10]
Lan
d
EH.
Rec
ent a
d
va
nces
i
n
R
e
tinex
the
o
r
y an
d so
me
in
pl
acatio
ns for
contica
l
co
mpu
t
ations c
o
l
o
r
vision
and th
e natura
l
i
m
ag
e
. Procee
din
g
of Natio
nal Aca
d
e
m
y
of Scie
nces
. 1983; 80(
16): 516
3-51
69.
[11]
DI XG, QU YY. An improve
d
lo
w
i
llum
i
n
a
ti
on im
age
en
h
ancem
ent al
go
rithm
w
i
t
h
col
o
r preservi
ng.
Journ
a
l of Har
b
in Inst
itute of
T
e
chno
logy.
2
014; 46(
3): 1-7
.
[12]
FAN YY, SHEN X
H
, SANG
YJ. No referenc
e im
ag
e sharp
ness ass
e
ssment base
d
on contrast
sensitivit
y.
Opt. Precisio
n Eng.
2011; 1
9
(10):
248
5-24
93.
[13]
XIE Z
H
X, LIU
YH, W
A
NG Z
H
F
,
XIONG XL,
HU Q. A nonli
near
l
y
c
o
mpe
n
s
ator
y
princ
i
pl
e
and meth
od
for human vis
i
o
n
contrast reso
lutio
n
.
Journ
a
l
of Chen
gd
u Medic
a
l Co
lle
ge.
2009; 4(3): 1
5
7
-16
2
.
[14]
CHEN Y, LI Y, Lü
XF
, XIE Z
X
, F
E
NG P. Active assessm
ent of col
o
r im
age
qua
lit
y ba
sed o
n
visu
a
l
perce
ption.
Op
tics and Precis
i
on Eng
i
n
eeri
n
g
.
2013; 21(
3): 2534-
254
2.
[15]
MA GF, YANG JH, ZHOU
B. Motion
d
e
tection
in
vi
de
o b
a
se
d o
n
Y
U
V co
lor s
pac
e.
Co
mput
e
r
Engi
neer
in
g an
d Desi
gn.
20
08
; 29(14): 37
00-
370
8.
[16]
MA YD,
T
E
NG F
,
Z
H
AN K, ZHANG HJ. A Ne
w
Meth
od o
f
Color Imag
e Enha
ncem
ent Using S
p
iki
n
g
Cortical Mo
de
l.
Journa
l of Beij
ing U
n
ivers
i
ty
of Posts and T
e
lec
o
mmun
icat
ions.
20
12; 35(
3): 70-73.
[17]
Z
H
AO XX, W
A
NG RL, Z
H
A
N
G LL.
Rese
a
r
ch
of
en
hanc
ement
alg
o
rith
m for d
egra
d
e
d
im
ages
i
n
advers
e
w
eat
h
e
r.
Mi
cro
c
om
pu
te
r Information
. 2010; 2
6
(8): 3-5.
[18]
LEE CH, S
H
I JL, LIEN
C
HCH, HA
N C
HCH.
Ad
aptiv
e Multisc
a
l
e
Retin
e
x for I
m
a
ge
Co
ntras
t
Enha
nce
m
ent
. Signa
l-Image
T
e
chnolog
y &
Internet-Base
d
S
y
stems, 201
3 Internatio
na
l
Confere
n
c
e
on IEEE. 2013:
43-50.
[19]
Lu Y, Guo
L, L
i
H. Rem
o
te
se
nsin
g ima
ge fu
sion
edg
e i
n
for
m
ation
an
d fea
t
ures of SAR i
m
age
bas
e
d
on curve
l
et transform.
Acta Photon
ica Si
nica
. 2012; 41(
9): 1118-
112
3.
[20]
Yong C
h
e
n
, Jie
Xio
ng, H
u
a
n
-lin
Liu, Qia
n
g
F
an. Comb
i
ne T
a
rget Ext
r
action a
nd E
nha
nceme
n
t
Methods
to F
u
se Infrare
d
a
n
d
L
LL Ima
ges.
T
E
LKOMNIKA Indon
esi
an J
o
urna
l of T
e
leco
mmu
n
icati
o
n
,
Co
mp
uting, El
ectronics a
nd
Contro
l).
2014,
12(3): 605-
61
2.
[21]
Jian
hua
L, Jia
ngu
o Y.
Multif
ocus im
age fu
sion
b
y
SML i
n
the sh
earl
e
t
subb
an
ds.
TEL
K
OMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
ng.
2014; 1
2
(1): 6
18-6
26.
Evaluation Warning : The document was created with Spire.PDF for Python.