TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 11, Novembe
r
2014, pp. 78
6
3
~ 786
8
DOI: 10.115
9
1
/telkomni
ka.
v
12i11.60
35
7863
Re
cei
v
ed Ma
rch 3
0
, 2014;
Re
vised Sept
em
ber
4, 201
4; Acce
pted
Septem
ber 2
0
, 2014
Color Difference Evaluation Model on Partly Changed
Complex Images
Zhang Zidon
g*, Jiemin Zhang, Yong
mei Li
Comp
uter Engi
neer
ing C
o
ll
eg
e, Jimei Un
iver
sit
y
(JMU)
No.18
3
Yinj
ia
n
g
Rd, 361
02
1 Jimei, Xiam
en,
F
u
jia
n, Chin
a, Ph./F
ax: +
86-5
92-6
182
45
1
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: zzd_88
8@a
l
i
y
u
n
.com
A
b
st
r
a
ct
Since th
ere
ha
s bee
n a stro
n
g
de
man
d
fro
m
i
n
d
u
stry to h
a
ve a
n
effici
en
t w
a
y of ma
na
gin
g
col
o
r
imag
e q
u
a
lity
prese
n
ted
in
d
i
fferent
med
i
a,
by sp
ecif
ica
lly
investi
gatin
g p
a
rtly
ch
an
ged
compl
e
x
i
m
ag
es,
this article pro
pose
d
a revisi
on to existin
g
CIE co
lor diffe
rence
mo
del w
h
ich ca
nnot gi
ve a prop
er col
o
r
differenc
e ass
e
ssment o
n
p
a
r
tly chan
ge
d c
o
mpl
e
x i
m
a
ges
. T
he key
met
hod
ap
pli
ed
is
to find
out w
e
i
ght
coefficie
n
ts of color attrib
utes
such as lig
htn
e
ss,
hue a
nd c
h
ro
ma i
n
color
differenc
e pre
d
i
ction.
Ke
y
w
ords
: col
o
r diffirenc
e, partly chan
ge
d compl
e
x imag
e, eval
uatio
n mod
e
l
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g an
d
Scien
ce. All righ
ts reser
ved
.
1. Introduc
tion
No
wad
a
ys co
lor imag
e re
p
r
odu
ction i
s
very com
m
on i
n
daily life. A
perfe
ct rep
r
o
ductio
n
sho
u
ld produ
ce an imag
e
identical to the origi
nal
o
ne. Practi
call
y, however, this is ra
rely the
ca
se d
ue to f
a
ctors
affecti
ng the final
i
m
age
of
rep
r
odu
ction, e.g.
device
gam
u
t, media ga
m
u
t,
sy
st
em n
o
is
e
,
sy
st
em e
rro
r,
et
c. Hence,
the rep
r
o
d
u
c
ed imag
e ma
y look q
u
ite d
i
fferent from t
h
e
origin
al
o
ne. To
d
e
ci
de wh
ether
the differen
c
e
i
s
n
o
ticea
b
le
or a
c
ceptable
is u
s
ually jud
ged
by
experie
nced
peopl
e. Ho
wever, for m
a
ssive reprodu
ct
ion, this be
comes unp
ra
ctical. The
r
efo
r
e,
there ha
s b
e
en a strong d
e
mand from
i
ndu
stry to ha
ve a metric th
at can qu
antify the differen
c
e
automatically by powe
r
ful compute
r
syst
em [1, 2].
Lots of effo
rt have be
en do
ne in the
pa
st
decade [3], h
o
weve
r, the i
m
age
s u
s
ed i
n
those
studie
s
were
mostly syst
ematic
ally
re
ndered, i.e. all pixels in
an imag
e were va
ried vi
a a
mathemati
c
al
function
on
a parti
cula
r
color attri
bute
su
ch a
s
lig
h
t
ness, ch
rom
a
, hue o
r
th
eir
combi
nation.
The intent
s
were to establish
perce
ptibility threshol
ds on different media and to
test the pe
rfo
r
man
c
e
s
of
color-differe
nce form
ul
ae.
Available research o
u
tco
m
e dem
on
strated
the failure of
conve
n
tional
formula
in
pre
d
icting
color
differen
c
e
s
fo
r two
ima
g
e
s
having
only p
a
rt
of the ima
ge
being
differe
nt. Additionall
y
in many
experim
ents, th
e ob
se
rvers
started
to
rep
o
rt
differen
c
e
bet
wee
n
two im
age
s
whe
n
o
n
ly pa
rt of
th
e differen
c
e
became
obvi
ous while th
e
re
st
still look
simi
lar. These
result
s suggest
that invest
igation of
partly
changed i
m
ages m
a
y
be
importa
nt as
well a
s
sy
ste
m
atically cha
nged o
n
e
s
. This a
r
ticle i
s
to de
sign a
n
experim
ent
to
investigate p
a
rtly cha
nge
d image
s an
d attempts
to
prop
ose an
initial model
that can fits
the
experim
ental
result. Further mo
re thi
s
mod
e
l
wo
uld be teste
d
by using t
he syste
m
ati
c
ally
changed images fo
r com
p
atibility.
2. Method
s
In the expe
ri
ment, re
sou
r
ce
s em
ploye
d
in
cl
ude
HP
p113
0 Mo
nitor, UNIX workstatio
n,
Photometer
PR650, M
a
tlab or
C/C++
and 1
4
hum
a
n
ob
serve
r
s. Thre
e main
stages
com
p
ri
sed
the experime
n
t, which ca
n be descri
bed as d
e
vice sim
u
latio
n
, subje
c
tive evaluation and
differen
c
e me
trics modelli
n
g
. The followi
ng se
ction p
r
ese
n
ts ea
ch
stage in mo
re
details.
2.1. Dev
i
ce Simulation
Device si
mu
lation o
r
d
e
vice
cha
r
a
c
teri
zation
model
enabl
es u
s
to t
e
st ou
r
unde
rsta
ndin
g
of the
devi
c
e
s
, predi
ct their
outpu
t,
and
optimize
their
de
sign.
As fa
r a
s
t
h
is
experim
ent i
s
co
ncerned,
we
only inve
stigate
CR
T
displ
a
y chara
c
teri
zation
m
odel [4],
whi
c
h
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 78
63 – 786
8
7864
covers gam
ma fun
c
tion,
sp
ect
r
al p
o
w
er di
st
rib
u
tion of
pho
sp
hors, an
d pi
xel point-sp
r
ead
function.
Con
s
id
erin
g t
w
o
CRT m
o
d
e
ls
availabl
e, GOG
a
nd P
L
CC [4], two
inde
pen
dent
testing
sets
are care
fully cho
s
en f
o
r thei
r evalu
a
tion. The
re
sults
by mea
n
⊿
E*ab b
e
tween a
c
tual va
lue
and it
s p
r
e
d
i
ction
usi
ng
GOG
and
P
L
CC m
odel
s
are li
sted i
n
Ta
ble
1;
Figure 1
is
its
corre
s
p
ondin
g
grap
hical re
pre
s
entatio
n.
As far a
s
this
experim
ent is con
c
e
r
ne
d, accura
cy is the main priori
ty for repres
enting the
transfo
rmatio
n betwe
en i
m
age pixel
RGB value
s
and its co
rrespon
ding t
r
istimul
u
s val
ues.
Although GO
G model ha
s the advantag
es of eas
y to impleme
n
t and few mea
s
urements, PL
CC
model requi
res la
rge n
u
m
ber of me
asu
r
eme
n
ts
but
yields bette
r accuracy. As seen
obviou
s
ly
from Figu
re1,
mean
⊿
E*ab
for PLCC all belo
w
one u
n
i
t, nearly half the value of GOG mo
del. So
PLCC m
odel
is justified to
cho
o
se for this experi
m
ent.
Table 1. Re
sult Data Sets
for GOG a
nd
PLCC
G
OG
PLCC
⊿
Eab EvenSpaced
1.46
0.87
Random
1.55
0.68
Average
1.51
0.77
Figure 1. Plots of Mean
⊿
E
*
ab from Eval
uating GO
G and PLCC M
odel
s
2.2. Subjectiv
e
Ev
a
l
uation
Followi
ng
ch
ara
c
teri
zatio
n
of the CRT
displa
y, su
bjective eval
uation is
co
ndu
cted,
who
s
e
goal
i
s
to fin
d
the
perce
ptibility thre
shol
d fo
r ea
ch
“a
re
a
ch
ang
e ratio” a
m
on
g pa
rtly
cha
nge
d ima
ges. Psy
c
ho
physi
cal jud
g
m
ents (det
e
c
tion, discrimi
nation an
d p
r
eferen
ce) a
r
e
made un
der
controlled vie
w
ing conditio
n
s (fixed lig
h
t
ing, viewing
distances, e
t
c) to gene
ra
te
highly relia
ble
and rep
eata
b
le data [5].
Reg
a
rdi
ng pa
rtly chang
ed
compl
e
x ima
ges, two ima
ges
comp
are
d
differ in col
o
r only for
a small p
a
rt. This exp
e
rim
ent is de
sign
ated ch
ang
e of individual o
b
ject (CIO) i
n
an image, which
differentiate
s from e
a
rlie
r studie
s
i
n
ve
stigat
ing
ch
a
nge of
entire
image
(CEI). Fou
r
ima
g
e
s
sho
w
n in Fi
g
u
re 2 a
r
e inv
e
stigate
d
. Since oth
e
r im
age
s we
re
condu
cted in
a simila
r ma
nner,
image1
_three
girl is taken
as an exam
p
l
e for the
followin
g
statem
ents of expla
nation, analy
s
is
and con
c
lu
sio
n
.
Figure 2. Fou
r
Test Imag
es used in the E
x
perime
n
t
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Colo
r Differe
nce Evaluatio
n Model on P
a
rtly Chan
ge
d Com
p
lex I
m
ages (Zh
a
n
g
Zidong
)
7865
Figure 3. Example of Choi
ce
for Obje
cts
in image1
_threegirl
The p
r
o
c
ed
ure for
rend
eri
ng pa
rtly ch
a
nged i
m
age
i
n
this exp
e
ri
ment is
expl
ained
as
follows
:
Firstly, cho
o
se important i
ndividual obj
ects in an im
age whi
c
h a
r
e to be chan
ged, the
wide
r the a
r
e
a
ch
ange
rati
o sp
rea
d
out
among
obje
c
t
s
, the more a
c
curate fin
a
l result
would
b
e
.
In Figure 3 si
x objects (are
a cha
nge ratio rang
e fr
om
2% to 6%) were isolated i
ndividually using
Photosh
op,
whi
c
h in
clu
d
e
thre
e gi
rls’
skin, top
ad
dre
s
ses an
d
blan
ket plu
s
gra
s
s a
s
a
whole
(area chan
ge
ratio about 1
9
%).
Secon
d
ly, sel
e
ct map
p
ing
function
s sho
w
n in Fig
u
re
4 to transfo
rm a particula
r col
o
r
dimen
s
ion
for ea
ch o
b
je
ct
in the o
r
igin
a
l
image,
su
ch
as
CIELAB
L*, C*
or
h.
Each tim
e
, o
n
ly
one pe
rceptu
a
l colo
r attrib
ute’s value
was modifie
d
in one obj
ect.
Figure 4. Tra
n
sfer F
u
n
c
tio
n
s u
s
ed in
CIO Experiment
The m
appi
n
g
fun
c
tion
s l
i
sted i
n
Fi
g
u
re
4
app
ro
ximate typical variatio
ns in
colo
r
rep
r
od
uctio
n
su
ch a
s
co
ntrast, gain, ga
mma co
ntrol
and color
shif
t, which a
r
e a
l
so u
s
ed in
CEI
experim
ent.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 78
63 – 786
8
7866
Figure 5. Wo
rk Flo
w
for P
a
rtly Chan
ge
d Images
Re
nderi
n
g
It is
c
r
itic
al to es
timate the two extreme le
vels whe
n
rend
eri
ng de
sire
d image
s:
One is
that more th
a
n
half pe
rcent
of the ob
se
rvers CA
NN
OT
see th
e difference; The
oth
e
r i
s
that mo
re
than half pe
rcent of the o
b
se
rv
ers CA
N dete
c
t the differen
c
e. Estimation for
the two extre
m
e
levels i
s
te
sted by a
grou
p of ob
se
rvers till t
he pi
np
oint levels are f
ound. T
h
u
s
the
perce
ptual
threshold fo
r
each obj
ect i
s
narro
wed
do
wn
within
the
above two lev
e
ls to find
out
. Choi
ce of th
e
in-bet
wee
n
le
vels ca
n be e
v
enly spa
c
ed
betwe
en the
above two ex
treme level
s
.
Thirdly, ren
d
e
r pa
rtly cha
nged ima
g
e
s
by m
eans of
C/C++ prog
ram [6]. Work flow o
f
image re
nde
ring pro
c
e
dure is illustrate
d in Figure 5,
where colo
r appe
ara
n
ce model refe
rri
ng to
CIELAB colo
r spa
c
e, the forwa
r
d CRT
model refe
rring to CRT chara
c
te
rizatio
n
model PLCC
whi
c
h u
s
e
s
th
ree l
ook-u
p table
s
rel
a
ting
digita
l count
s an
d radiom
etric
scala
r
s f
o
r e
a
ch chan
nel,
CIE spe
c
ifica
t
ion are obtai
ned thro
ugh
a linear tr
a
n
sformation bet
wee
n
radi
om
etric scal
ars and
tris
timulus
values
.
The psy
c
ho
p
h
ysical judgm
ents we
re co
ndu
ct
ed on a
HP p1130 m
onitor in a da
rk room.
Before ea
ch
ob
serve
r
’s observation, device
cali
bration
was ad
justed
to en
sure the
mo
nitor
perfo
rm
in a pred
efined way
(ad
opted white
p
o
int,
c
ontra
st, brig
h
t
ness, etc) in
whi
c
h
con
d
itions
CRT
di
splay
cha
r
a
c
teri
zati
on
wa
s d
e
vel
oped
[7]. A p
a
ir
of imag
es wa
s present
ed
side
by
si
de
each time o
n
monitor
and
the origi
nal i
m
age i
s
rand
omly positio
n
ed on l
e
ft or
right to mini
m
i
ze
the ne
gative
effect of m
oni
tor‘s spatial u
n
iformi
ty. Ob
serve
r
s
were
asked
to ju
dg
e whethe
r th
ey
coul
d see a
differen
c
e or not between
two im
age
s, and the obj
ect exhibiting
difference was
spotted fo
r validity following the detectio
n
of difference.
3. Results a
nd Analy
s
is
Obje
ctive in the follo
wing
work i
s
to det
ermin
e
fun
c
tions
between
threshold
⊿
E and
a
r
e
a
cha
nge ratio, the following
gives its explanat
io
n: Firstly, relationship betwe
en
obje
c
t mean
⊿
E
and color attri
bute [L, C, h]
is prove
d
to
be linear a
c
co
rding to ded
uced Equation
(1):
0
0
0
*
*
*
2
2
2
*
C
L
when
h
ab
E
h
L
when
C
ab
E
h
C
when
L
ab
E
h
C
L
ab
E
(1)
Therefore, fu
nction
(de
not
ed by f ) a
s
sociatin
g col
o
r attribute [L, C or
h] and d
i
fference
detectio
n
rati
o by obse
r
vers ca
n tran
sformed
to functi
on (de
noted
by f’) relating obje
c
t mean
⊿
E
and diffe
ren
c
e dete
c
tion
ratio by ob
se
rvers.
Whi
c
h
can be
sim
p
lified a
s
f{x, y}
→
f’{X, y}, where
x stand
s for
one
colo
r attribute [L, C
or h], X equal
s
to obje
c
t mea
n
⊿
E, y repre
s
ent
s differen
c
e
detectio
n
rati
o by obse
r
vers.
Secon
d
ly, by fitting the result data sets th
rough m
a
thematical too
l
s, function f’
can be
obtaine
d. In succe
ssi
on, va
lue of th
re
sho
l
d
⊿
E for e
a
ch
are
a
cha
nge
ratio i
s
fixed
on by
setting
y
equal
s to 0.
5. Finally, function
reg
a
rd
ing thre
sh
old
⊿
E and corresp
ondi
ng a
r
ea chang
e ratio
usin
g re
sult d
a
ta sets from the above fo
u
r
image
s is ill
ustrate
d
in Fi
gure 6.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Colo
r Differe
nce Evaluatio
n Model on P
a
rtly Chan
ge
d Com
p
lex I
m
ages (Zh
a
n
g
Zidong
)
7867
Figure 6
.
Plo
t
Using F
our I
m
age
s’ Re
sul
t
Data Sets b
e
twee
n Area
Cha
nge
Ratio
s
and O
b
je
ct
Thre
sh
old Me
an
⊿
E
In Figu
re 6
it is a
ppa
re
nt that the lig
htn
e
ss differen
c
e (d
enote
d
b
y
blue p
o
ints) is le
ss
notice
able th
an ch
rom
a
a
nd hue differences (den
oted by yellow and red p
o
i
n
ts re
spe
c
tively)
whe
n
evalu
a
te imag
e diffe
ren
c
e
on p
a
rt
ly chan
ged
complex ima
g
e
s. In a
dditio
n
, the correla
t
ion
betwe
en
are
a
ch
ang
e ratio
and
co
rrespo
nding
thre
sh
o
l
d obj
ect m
e
a
n
⊿
E is
currently still low, the
rea
s
on
s for th
at may be explaine
d as foll
ows:
One i
s
that threshold
⊿
E al
so d
epen
ds
o
n
factors
su
ch as
whi
c
h o
b
ject, its po
si
tion and
how la
rge a
r
e
a
wa
s sel
e
cte
d
, etc. Theref
ore, more
factors ne
ed to b
e
taken into
consi
deration to
maximize the
correlation b
e
twee
n imag
e threshold
⊿
E and area
chang
e ratio; Another p
r
obl
em
is that wheth
e
r the
experi
m
ent
re
sult i
s
image d
epe
ndent, in oth
e
r words,
can
it be gen
eral
ize
d
to other type of images?
Therefore diff
erent ty
pe of images n
e
e
d
to be investigated to see
wheth
e
r the furthe
r re
sults
obtaine
d sh
o
w
agreem
ent with earli
er o
nes, that is th
e function
curve
betwe
en im
a
ge threshold
⊿
E and a
r
e
a
ch
ang
e ra
tio; Finally, the fou
r
ima
ges
used i
n
the
experim
ent a
r
e of th
e
sa
me si
ze
an
d
image
si
ze i
s
al
so
a fa
ctor affe
cting i
m
age
differe
nce.
Thereby, wh
ether the im
age si
ze
will have an influ
ence on the
experim
ental
result is still
an
open q
u
e
s
tio
n
.
In a
wo
rd, th
e obj
ective
is to
combi
n
e
all countin
g f
a
ctors to g
e
t the
right
fun
c
tion
F
illustrate
d in Figure 7, whi
c
h rel
a
ting im
age thre
sh
old
⊿
E and are
a
cha
nge ratio.
Figure 7. Fun
c
tion Curve Relating Th
re
shold
⊿
E and
Area Chan
ge
Ratio
4. Conclusio
n
Since
colo
r d
i
fference is th
e main difference betwee
n
image
s, so
CIELAB formula wa
s
adopte
d
for revising
thro
u
gh a
dding
weight
coeffici
ents
derive
d
from thi
s
exp
e
rime
nt on
color
attribute [L,C or h]. In Eq
uation (2
) be
low, if the
⊿
E*ab_
CIO is great than
one unit, image
differen
c
e sh
ould be
n
o
ticeable, whe
r
e
weig
ht
coe
fficient (L
w, L
w
,
hw) can
be
o
b
tained
thro
u
g
h
function Y =
F (X) in Figu
re 7.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 11, Novem
ber 20
14: 78
63 – 786
8
7868
hw
h
Cw
C
Lw
L
CIO
ab
E
2
2
2
*
_
(2)
Our ultim
a
te goal is to i
d
e
n
tify the visual
sign
als th
a
t
annoy ou
r viewin
g cu
sto
m
ers and
minimize thei
r visibility. Device
simul
a
tions ena
ble
us to isolate source
of a visual
signal.
Psycho
physi
cal evaluatio
n
enabl
es
us to
asse
ss rol
e
of the si
gnal
plays in
su
bje
c
tive judg
me
nts
of image
qu
al
ity. Eventually, differen
c
e
metrics
help
us to
qu
antify or mea
s
u
r
e
the ma
gnitud
e
of
the sign
al.
Furthe
rmo
r
e,
it would
be in
terestin
g to te
st
S-CIELAB
[8] using th
e
CIO expe
rim
ent data
sets an
d
wh
at are the
p
r
edictio
n
resul
t
s fr
om
S-CI
ELAB for
evaluating
ima
ge diffe
ren
c
e
on
partly cha
nge
d compl
e
x image
s wo
uld b
e
.
Referen
ces
[1]
F
a
rrell JE.
Ima
ge
qu
alit
y
eval
uatio
n. C
o
lor
Imagi
ng:
Vis
i
o
n
an
d T
e
chno
lo
g
y
.
Joh
n
W
i
l
e
y & So
ns
Ltd.
199
6.
[2]
F
a
rrell JE, Silv
erstein DA, Z
h
ang
X.
Ima
ge
qua
lity metrics base
d
on si
ngl
e and
mu
lti-ch
ann
el
mod
e
l
s
of visual pr
oce
ssing
. Proceedings of IEEE Compcon 97. 1997; 57-60.
[3]
Silverstein
DA, Farrell JE.
Qu
antifyin
g
perce
ptual
q
ual
ity.
Proc. 51st IST
Annu
al M
eeti
ng,
19
98;
8:81
-
84.
[4]
Berns RS. Methods for char
a
c
terizin
g
CRT
displ
a
ys. Elsev
i
er. 199
6; 16(4)
: 173–1
82.
[5]
F
o
le
y JM,
Leg
ge GE. C
ontra
st detectio
n
an
d n
ear-thres
ho
ld
discrimi
nati
o
n i
n
h
u
man
vi
sion.
V
i
sio
n
Res.
198
1; 104
1-53.
[6]
Pratt W
K
. Digital Image Proc
e
ssing, 2n
d. Ne
w
York: Jo
hn
W
ile
y
.
1
9
9
1
.
[7]
Brain
a
rd DH. C
a
libr
a
tio
n
of a computer-c
ontro
lled c
o
lor mo
nit
o
r.
Color R
e
s. Appl.
19
84; 14:
23-3
4
.
[8]
Xu
eme
i
Z
han
g, Brian A.
W
ande
ll. A Spatia
l
Ext
ensi
on of CIELA
B
for Digita
l
Color Ima
g
e
Repr
oducti
on.
SID Digest.
19
96; (7): 31-34.
[9]
Dal
y
SJ. T
he visibl
e differ
e
n
c
es pred
ictor: an al
gor
ithm for the assess
ment of
imag
e
fidelit
y. Di
gital
Images an
d Hu
man Visi
on. Ca
mbridg
e MA: MIT
press. 1992.
[10]
Carlso
n CR,
Coh
en RW.
A simple
psycho-phys
i
cal
m
o
del for pr
edicting the vis
i
bilit
y of displaye
d
infor
m
ati
on.
Proc. SID. 1980; 21(3): 22
9-4
6
.
[11]
Sn
yder H
L
. Image q
u
a
lit
y
an
d observ
e
r per
formanc
e, in B
i
berma
n, L. (ed.
) Perceptio
n of Displ
a
y
e
d
Information. N
e
w
Y
o
rk: Plen
u
m
Press. 1973.
[12]
Silverstei
n DA,
F
a
rrell JE.
T
h
e
rel
a
tio
n
shi
p
betw
een i
m
ag
e
fide
lity
a
nd i
m
a
ge qu
ality.
Proc,
ICIP-96.
199
6; 881-
84.
[13]
Sharma G, W
u
W
,
Dala
l E
N
. T
he CIEDE200
0
co
lor
differenc
e for
m
ula: Impl
em
entatio
n n
o
tes
,
supp
leme
ntar
y test data, a
n
d
mathematic
al
observ
a
tions
. Color Res
earc
h
&
App
lic
ation
. 2005;
30(1)
:
21-3
0
.
[14]
W
u
Y, Z
hao XP, Jin Y, et al.
Extractin
g
Gol
den Are
a
from Image Base
d o
n
Otsu Algorith
m
[J]. Applied
Mecha
n
ics an
d
Materials. 20
1
4
; 469: 26
0-26
4.
[15]
Li Z
J
, Men
g
Q
J
. Rese
arch
on
the C
o
l
o
r-Diffe
r
ence
Eval
uati
on
of F
ood
Pa
ckage
Printi
ng
Matter Base
d
on the Hum
an
Visua
l
Char
acteristics. Appl
ie
d
Mecha
n
ics a
nd Materi
als. 2
014; 46
9: 278-
281.
[16]
Rasma
na ST
,
Supra
p
to YK, Purnam
a KE.
A Study of Col
o
r Differenc
es
on the Met
a
l In
scriptio
n
Ima
g
e
Based o
n
CIELab C
o
lor Sp
ace
. Semin
a
r on Intele
ge
nt T
e
chnolog
y a
nd Its Applica
t
ion (SIT
IA).
Surab
a
y
a, Indo
nesi
a
. 201
3.
[17]
Mal
y
khi
n
a
MP, Shichk
in
DA.
Aspects of P
r
actical
Use O
f
Color
Differe
nce F
o
r
Rec
o
gniti
on A
n
d
Selecti
on of
Boun
dar
y
Lin
e
On Images.
Polythe
m
atic Onlin
e
Scie
ntif
ic Jour
nal
of Kuba
n Stat
e
Agrarian University
. 2013.
[18]
Shame
y
R, Li
n J, Sa
w
a
t
w
a
r
akul W
,
et al.
Evaluati
on of
performanc
e of various co
l
o
r
differenc
e
formula
e usin
g
an e
x
per
iment
al bl
ack datas
e
t.
Color Rese
ar
ch & Applic
atio
n.
2013.
[19]
Kishi
no M, M
i
yaz
a
ki
S. Ima
ge
process
i
n
g
metho
d
, ima
ge
process
i
n
g
ap
parat
us, a
nd
ge
nerati
n
g
method: U.S. Patent Appl
icati
on 13/7
94,3
26[
P]. 2013.
[20]
Liu H, H
u
a
ng
M, Liu Y, et a
l
.
Color
Differe
nce Eva
l
uati
o
n
and C
a
lc
ulati
on for Di
gita
l and Pr
int
e
d
Imag
es
. NIP & Digital F
abr
ic
ation C
onfere
n
c
e. Soci
et
y
for
Imaging Sci
e
nce an
d T
e
chnol
og
y. 201
2;
(1): 140-1
43.
[21]
Qu Z
H
. An Algorithm of Im
age Qu
alit
y A
sse
ssment Ba
sed o
n
Data
F
i
tting of Image Histo
gram
.
T
E
LKOMNIKA Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
n
g
.
2014; (12): 5
9
9
-60
7
.
[22] Milad
Shirm
o
h
a
mmadi. Sta
n
d
a
rd
izi
ng S
egm
entin
g a
nd T
e
n
derizi
ng
Letter
s
and Impr
ovi
n
g the Qu
alit
y
of Envel
ope I
m
ages to E
x
tr
act Postal Add
r
esses.
Interna
t
iona
l Jour
nal
of Electrical
a
nd Co
mpute
r
Engi
neer
in
g
. 2012; (2): 38
3-3
94.
Evaluation Warning : The document was created with Spire.PDF for Python.