TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.5, May 2014, pp
. 4107 ~ 41
1
4
DOI: http://dx.doi.org/10.11591/telkomni
ka.v12i5.4228
4107
Re
cei
v
ed O
c
t
ober 1
4
, 201
3; Revi
se
d Decem
b
e
r
29, 2013; Accept
ed Ja
nua
ry 2
0
, 2014
An Ada
p
tive All-odd Transformation Watermark
Scheme
Qiaomei Ma, Lijun Wu*, Jianhong Du, Gouxi Ch
en,
Qiuxiang Ya
ng
Comp
uter and
Contro
l Engi
ne
erin
g, North Un
iversit
y
of Chi
n
a,
T
a
i
y
uan 0
3
0
051, Ch
in
a
*
Corres
p
o
ndi
n
g
author, e-ma
i
l
: LIJUNW
U07
07@
163.com
A
b
st
r
a
ct
Com
b
ining dis
c
rete
wavelet
transfor
m
, human visual
system
, chaotic map
and odd-ev
en
qua
nti
z
a
t
i
on, a nov
el w
a
ter
m
ark
i
ng sc
he
me for
2D i
m
a
ges is
pro
pose
d
in th
is
pap
er. Encrypte
d
w
a
termarki
ng and ide
n
tity
inf
o
rmatio
n
are i
m
p
e
rce
p
tibly
e
m
b
e
d
ded
in
2D
imag
es accor
d
ing to th
is sche
m
e.
F
o
r improvi
ng
the accur
a
cy o
f
just
noticea
bl
e distortio
n
a
n
d
enh
anc
e ro
b
u
stness of al
g
o
rith
m, the mo
del
ada
ptively
adj
usts the thres
hol
d inte
nsity
by ad
din
g
w
e
i
ghtin
g factor a
nd stren
g
the
n
i
ng the
brig
htn
e
ss
sensitivity ca
lc
ulati
on
meth
od
of ju
st noticea
ble d
i
stortion.
T
he bij
e
cti
on r
e
lati
onsh
i
p is c
onstructed b
e
t
w
een
the i
m
age
co
efficients
and
w
a
termarki
ng
bit
s
by th
e w
a
y
of al
l-od
d tra
n
s
formati
o
n
to the
just n
o
tice
a
b
le
distortio
n
, so t
he sch
e
m
e r
e
ali
z
e
s
bl
ind
ex
traction of w
a
t
e
rmark. T
he e
x
peri
m
e
n
tal r
e
sults an
d a
nal
ysis
show
that the
prop
osed sc
he
me c
an res
i
st intenti
ona
l
or uni
ntent
i
o
n
a
l vari
ety of attacks and
h
a
v
e
super
iority co
mpari
ng w
i
th exi
s
ting sche
m
es.
Ke
y
w
ords
:
di
g
i
tal w
a
termarki
ng, just notic
ea
ble d
i
st
ortion, l
ogistic
ma
ppi
n
g
,
z
i
g
z
a
g
sca
n
n
in
g
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
Digital
wate
rmarking i
s
a
comm
uni
cat
i
on p
r
oble
m
with si
de inf
o
rmatio
n, na
mely a
watermarkin
g
system which com
p
letely boycott t
he i
n
terferen
ce b
e
twee
n ho
st sign
al and hi
de
sign
al [1]. For getting a
bet
ter visu
al effe
ct, lo
ts of
wo
rk have
bee
n
dev
oted to
st
udying the
HVS
(hum
an visu
al model). Z
hou [2] etc. establi
s
h
ed
a
MJND mod
e
l(ju
st notice
able disto
r
tio
n
in
Multiview) for the study of
HVS to stere
o
scopi
c
mas
k
ing effec
t. Literature [3-6]
pres
ented s
o
me
algorith
m
s which
ba
sed
o
n
the
DCT
(Discrete
Co
sin
e
Tran
sform
)
comp
re
ssible
Wat
s
on
mod
e
l.
Afterwa
r
ds, t
he Barni p
e
rceptual m
odel
[7-11]
b
a
sed
on DWT
(Discrete
Wavelet
Tra
n
sfo
r
m) h
a
s
been
widely u
s
ed in di
gital watermarkin
g
.
Fang [10] an
d Huan
g [11] etc. had improved
o
n
Barni mod
e
l. They use
d
the Barni
model to
dete
r
mine
the e
m
beddi
ng
stre
n
g
th. Fang
ch
ose
a
stro
ng
robu
stne
ss
cofficient throu
g
h
the analysi
s
of the waterm
arki
ng
capaci
ty as the embedding bit,ther
eby the imperceptivity of
the
watermark
was improved. Huan
g described the probability den
sity function of wavelet
coeffici
ents
b
y
using the
G
eneralized G
aussia
n
Di
stri
bution (GG
D
) to balan
ce th
e impe
rceptivity
and robu
stne
ss. Be
ca
use
embed
ding l
o
cation
and
cover ima
ge a
r
e ne
ede
d in
the wate
rma
r
k
extraction, th
eirs
algo
rith
ms have
hig
h
com
p
le
xity and
can’t a
c
hieve th
e bl
ind extra
c
tio
n
of
water
m
ar
k.
In ord
e
r to
overcome t
h
e above
sho
r
tcomi
n
g
s
, a
n
ada
ptive a
ll-odd
tran
sfo
r
mation
watermarkin
g
sche
me i
s
propo
sed.
On
the b
a
si
s
of
th
e Barni m
ode
l, usin
g th
e lu
minan
ce
fact
or
and wei
ght factor to opti
m
ize the JND for improv
i
ng the se
curi
ty of the waterma
r
k.M
o
re
over,a
blind extra
c
ti
on of wate
rm
ark
algo
rithm
base
d
on
JND i
s
p
r
opo
sed. The
sche
me not only
has
better impe
rceptivity and robu
stne
ss, b
u
t also is
con
v
enient and h
i
ghly efficient.
The
re
st of t
he p
ape
r i
s
orga
nized
as follows. Th
e
pro
p
o
s
ed
waterma
r
ki
ng
scheme,
optimizin
g threshold, an
al
yzing the all
-
odd
tran
sformation, the watermark e
m
beddi
ng an
d
extract-i
on a
r
e d
e
tailed i
n
Sectio
n 2.
Experime
n
tal re
sult
s an
d pe
rform
a
n
c
e a
nalyses are
discu
s
sed in
Section 3, an
d the co
n
c
lu
si
on is drawn in Section 4.
2. Proposed
Wa
termar
king Scheme
2.1. Model Ov
er
v
i
e
w
Figure 1 sh
o
w
s the p
r
op
o
s
ed mo
del of watermarkin
g
sch
eme.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4107 – 41
14
4108
Figure 1. Model of Wate
rmarking Sche
me
The mod
e
l of watermarkin
g
sch
eme
con
s
ist
s
of three
module
s
de
scrib
ed a
s
follows:
(1) Cover p
r
etreatme
nt: Any cove
r i
s
processe
d b
y
frequ
en
cy-domain
tra
n
sform a
n
d
matrix of freq
uen
cy coeffici
ent is acquire
d.
(2) Waterm
ark p
r
etreatme
nt: In the
pro
c
e
ss,
waterm
ark i
s
scann
e
d
an
d
scram
b
led to
get
the final emb
edde
d se
que
nce.
(3) Embed
din
g
and
extra
c
ti
ng
watermark: As the i
n
termediate
porti
on of Fi
gure1
. Whe
n
watermark is embedd
ed, the cove
r freq
uen
cy-do
m
ai
n coeffici
ents after pretre
a
t
ment are firstly
pro
c
e
s
sed by
HVS model and the JND is acqui
re
d. The JND d
e
termin
es em
b
edde
d stre
ng
th.
The en
crypte
d wate
rma
r
k is emb
edd
e
d
into
freq
u
ency-dom
ain
coeffici
ents using
odd
-e
ven
analysi
s
, an
d
the wate
rma
r
ke
d ele
m
ent
are g
ene
rate
d finally. The
watermark
can be
extract
ed
throug
h freq
u
ency-dom
ain
transfo
rmin
g, odd-even an
alyzing a
nd a
n
ti-scramblin
g of the wate
r-
marked el
em
ent.
Model fra
m
e
w
ork i
s
de
scribed a
s
an ei
ght-ele
ment
array:
F
=
f
(
C
,
W
,
C
′
,
W
′
,
EN
,
K
,
EM
,
EX
). The detailed de
scripti
ons a
r
e a
s
follows:
The set of co
ver
i
s
d
e
fined
as
C
={
C
1
、
C
2
…
C
r
…}.
D
1
、
D
2
…
D
n
are th
e fre
que
ncy-domain
coeffici
ents correspon
ding to
the
cover
C
r
.
W
={
W
1
、
W
2
…
W
l
…}, the s
e
t of watermark
.
C
′
={
C
′
1
、
C
′
2
…
C
′
r
…}, the
se
t of waterm
arked
eleme
n
ts.
D
′
1
、
D
′
2
…
D
′
n
are the f
r
eq
uen
cy-
domain
coeffi
cient
s co
rrespondi
ng to the watermarke
d
C
′
r
.
W
′
=
{
W
′
1
、
W
′
2
…
W
′
l
…},the
set of
watermark p
r
o
c
e
s
sed
by
en
cryption which
ca
n
be
embed
ded in
cover.
EN
r
e
pr
es
ents
w
a
ter
m
ar
kin
g
e
n
c
r
yp
tion
a
l
g
o
r
i
th
m. W
a
te
rma
r
k
W
l
is
scan
ned an
d
scram
b
led
be
fore
embe
dd
ed into
the
cover,so
that
t
he
se
cu
rity of the
wate
rma
r
k is en
han
ce
d.
Variabl
e
K
n
eede
d in
wat
e
rma
r
k extra
c
tion i
s
th
e key gen
erate
d
at the time
of the
watermark
scram
b
ling. T
he co
rrespon
ding fun
c
tion
is (
W
′
,K
)=
EN
(
W
).
EM
represent
s wate
rma
r
ki
ng embe
ddin
g
algorith
m
. The functio
n
is
C
′
=
EM
(
C
,
W
′
).
EX
is the waterma
r
k ext
r
actin
g
algo
ri
thm which is used to extract waterm
a
r
k fro
m
watermarke
d element
s to prove the pr
od
uct co
pyright,
e
tc. The fun
c
tion is
W
=
EX
(
C
′
,
K
).
2.2. Model Analy
s
is
2.2.1. JND P
a
rameters O
p
timization
JND
refers t
o
the minim
u
m disto
r
tion t
hat
ca
n b
e
i
dentified in t
he expe
rime
nt, which
reflect
the
p
e
rception
ch
ara
c
teri
stics
of HVS di
re
ctly [5]. There
f
ore,
sho
w
n
from the
mo
del
overview, the
acqui
sition a
nd pro
c
e
s
sin
g
of JND is th
e key part of the algo
rithm.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Adaptive
All-odd T
r
an
sform
a
tion Wa
term
ark Sche
m
e
(Qiaom
ei Ma)
4109
Barni’
s [8] definition abo
ut JND main
ly
consi
d
e
r
e
d
the frequ
e
n
cy se
nsitivity
F
(
l
,
θ
),
luminance sensitivity
L
(
l
,
x
,
y
), textured
and
edg
e
cha
r
a
c
teri
stics
T
(
l
,
x
,
y
). T
he value
is
the
weig
hted re
sult of three terms a
bove an
d the basi
c
e
quation
s
are as follo
ws:
2
.
0
)
,
,
(
)
,
,
(
)
,
(
2
1
)
,
(
y
x
l
T
y
x
l
L
θ
l
F
y
x
JND
θ
l
(
1
)
3
2
1
0
,
10
.
0
,
16
.
0
,
32
.
0
,
00
.
1
1
,
,
1
2
)
,
(
l
l
l
l
if
if
if
if
otherwise
θ
if
θ
l
F
(
2
)
otherwise
A
A
if
A
y
x
l
L
,
1
5
.
0
,
2
)
,
,
(
,
)
2
1
,
2
1
(
256
1
3
3
3
3
l
l
y
x
D
A
(3)
1
,
0
,
3
3
3
3
2
3
0
2
0
1
0
1
0
)
2
1
,
2
1
(
)]
2
,
2
(
[
16
1
)
,
,
(
j
i
l
l
l
k
θ
ij
k
k
θ
l
k
k
y
j
x
i
D
Var
y
j
x
i
D
y
x
l
T
(
4
)
Whe
re
l
is th
e sub
-
b
and a
t
decom
po
sition level,
θ
∈
(0,1,2,3) re
pre
s
ent
s high fre
quen
cy,
hori
z
ontal,ve
r
tical an
d lo
w f
r
equ
en
cy co
mpone
nt afte
r tra
n
sfo
r
min
g
,
D
θ
l
(
x
,
y
)
is
th
e
c
o
rr
es
p
ond
i
n
g
coeffici
ent of coo
r
din
a
te (
x
,
y
).
Shown in Eq
uation (1
)-(4
), the JND whi
c
h corr
e
s
po
n
d
s to co
effici
ent at the sa
me place
is
same
in ev
ery level of t
he
sub
-
ba
nd
s. In o
r
de
r to
eliminate
th
e problem,
combine
with
HVS
prefe
r
ably an
d enha
nce the algorith
m
's
robu
stne
ss
, this pa
per m
a
ke
s improve
m
ent as follo
ws:
(1)
Cha
nging th
e
luminan
ce fa
ctor
For
red
u
ci
ng
the compl
e
x
ity of the algorithm, th
e
DWT
coeffi
cient
s i
s
no
rmalize
d
.
Equation (3) i
s
improved a
s
follows:
otherwise
A
A
ifA
A
y
x
l
L
,
5
.
1
,
2
)
,
,
(
(
5
)
Ā
is the mea
n
of
A
. Equation (5)
can
be better
cap
t
ure the eye
brightn
e
ss se
nsitivity
becau
se it ca
n obtain the b
e
st matching
with different
covers that
Ā
is ch
osen.
(2) Addi
ng a
weig
hting fact
or
As the
coefficients
co
rresp
ondin
g
to
different
po
sition
s a
r
e
different
, the
weig
htin
g facto
r
is define
d
by its own attri
b
u
t
e:
)
,
(
)
,
(
y
x
D
y
x
D
α
θ
l
θ
l
,
)
,
(
y
x
D
θ
l
is the mea
n
of each sub-band
coeffici
ent.
The value
of
threshold
s
ca
n be a
daptiv
ely cha
nge
d
by addin
g
a
weig
hting fa
ctor so a
s
to
adjust
the embed
ding strength.
From all of ab
ove, Equation
(1) is
redefin
ed as:
2
.
0
)
,
,
(
)
,
,
(
)
,
(
2
)
,
(
y
x
l
T
y
x
l
L
θ
l
F
α
y
x
JND
θ
l
(
6
)
2.2.2. All-od
d Trans
f
orm
a
tion
“All-od
d tra
n
sformation
” is
an adj
ustme
n
t to
JND fo
r maki
ng the
threshold i
n
to "odd",
whi
c
h
aims to con
s
tru
c
t a
bije
ction
rela
tionshi
p b
e
tween th
e ima
g
e
coefficie
n
ts and
waterm
ark-
ing bits, reali
z
es blin
d extra
c
tion of
wate
rmark. The eq
uation is a
s
follows:
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4107 – 41
14
4110
If
)
10
)
,
(
(
y
x
JND
round
θ
l
is even, then
10
/
)
1
)
10
)
,
(
(
(
)
,
(
y
x
JND
round
y
x
Q
θ
l
θ
l
(
7
)
Q
θ
l
(
x
,
y
) i
s
t
he
corre
s
p
o
nding
thre
sh
old of
co
ordinate
(
x
,
y
).
For em
bed
ding th
e
watermark by
the odd-eve
n
modulation
method, we l
e
t:
1
)
1
(
0
)
1
(
1
0
w
and
even
U
Q
U
w
and
odd
U
Q
U
w
and
odd
U
or
w
and
even
U
Q
U
D
θ
l
θ
l
θ
l
θ
l
θ
l
θ
l
θ
l
θ
l
θ
l
θ
l
θ
l
,
]
)
,
(
)
,
(
[
y
x
Q
y
x
D
round
U
θ
l
θ
l
θ
l
(8)
Whe
rein,
D'
repre
s
e
n
t
the watermarke
d coeffici
ent
an
d
w'
i
s
the
wa
termarkin
g
bit
.
Since
there
is onl
y the
watermarked
info
rmation
wh
en
wate
rma
rk dete
c
tion, t
h
rou
g
h
reve
rse
dedu
ction, th
e prin
ciple of
watermark bli
nd extractio
n
is as follo
ws:
If
U
θ
l
=2N,
D
'
θ
l
=
U
θ
l
·
Q
θ
l
=2N,
then w'=0,
or
D
'
θ
l
=(
U
θ
l
+1
)·
Q
θ
l
=2N+1,
then
w'=1.
If
U
θ
l
=2N
+
1,
D
θ
l
'=
U
θ
l
·
Q
θ
l
=2N+1,
then
w'=1,
or
D
'
θ
l
=(
U
θ
l
-1
)·
Q
θ
l
=2N,
then
w'=0.
Thus th
e waterma
r
ki
ng
informatio
n can b
e
dire
ctly obtaine
d by the parity of
watermarke
d image coeffici
ents.
2.3. Watermark Embeddi
ng and Extr
a
c
tion
For some DWT co
efficien
ts of the intermediate an
d high freq
uen
cy sub-b
and
s are 0 o
r
negative, DW
T coefficie
n
ts may be overflow wh
e
n
wa
termark is e
m
bedd
ed. Fo
r above re
ason,
the 4-th level
low-frequ
en
cy sub-b
and i
s
sele
ct
ed to be embe
dde
d watermark.
The ba
sic
steps
are:
Step 1. Wate
r
mark p
r
etre
atment: Binary image is
scan
ned u
s
in
g
Zigza
g
meth
od and
redu
ce
d dim
ensi
on. The
scann
ed results
W
i
,
i
∈
(1
,2…
m
) are
scra
mble
d by
the Logi
stic
map
(
x
0
=0.2589), Setting
key
K
=
x
0
. Finally,
watermarking
W
′
(
i
,
j
),
i
×
j
∈
(1,2…
m
) can b
e
obtaine
d from
W
i
by the Zigzag a
n
ti-sca
n
n
ing.
Step 2. Calcu
l
ation of JND
and thre
sh
old
:
Cover imag
e is de
comp
o
s
ed to four le
vels to
get the coeffi
cient
s
D
θ
l
(
x
,
y
).Com
pute th
e
JN
D
3
3
(
x
,
y
) and
Q
(
x
,
y
),
x
∈
(1,2…
M
/16),
y
∈
(1,2…
N
/16)
throug
h the e
quation
s
in Section
2
re
spe
c
tively.
Step
3.
Embe
dding wate
rm
ark: The co
efficient
of
cu
rre
n
t embe
dde
d
locatio
n
is mo
dulat-
ed throu
gh th
e Equation
(
8).
Step 4.
Gene
rating
the
wa
termarke
d im
age: T
he i
m
a
ge
with
embe
dded
waterm
ark
wil
l
be gen
erate
d
after IDWT.
Step
5.
Wa
termark extraction and decrypt
ion:Ba
sed
on the
Step 2, waterm
a
rk
extraction
is
quick a
nd
ea
sy, whi
c
h i
s
makin
g
a
jud
g
ment fo
r pa
rity of co
effici
ents
after
DWT.
Extraction eq
uation is:
odd
is
even
is
D
D
w
i
1
0
(
9
)
The re
sult
s are p
r
o
c
e
s
se
d by decryption of cha
o
tic sequ
en
ce u
s
ing the
key
K, and
obtaine
d the watermark inf
o
rmatio
n.
3. The Exper
i
mental Res
u
lts and Ana
l
y
s
is
3.1. Ev
aluation Crite
r
ion and Ba
sic Experiment Si
mulation
The p
e
a
k
sig
nal-to
-
noi
se
ratio (PSNR)
and
stru
ctu
r
al
simila
rity(SSIM) a
r
e th
e ev
aluation
of qualities o
f
waterma
r
ke
d image. No
rmalize
d
co
rrelation (NC)
and bit error
ratio (BER) a
r
e
use
d
as the e
x
tract wate
rm
ark p
e
rfo
r
ma
nce in
dex, T repre
s
e
n
ts the
elapsed time
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Adaptive
All-odd T
r
an
sform
a
tion Wa
term
ark Sche
m
e (Qiaom
ei Ma)
4111
]
)]
,
(
)
,
(
[
255
lg[
10
11
2
2
M
x
N
y
y
x
D
y
x
D
N
M
PSNR
(
1
0
)
)]
,
(
[
)]
,
(
[
)]
,
(
[
)
,
(
D
D
s
D
D
c
D
D
l
D
D
SSIM
(
1
1
)
j
i
j
i
j
i
j
i
w
j
i
w
j
i
w
j
i
w
NC
,
2
,
2
,
)
,
(
)
,
(
)
,
(
)
,
(
(
1
2
)
j
i
w
w
w
w
mn
w
w
BER
,
,
0
,
1
100
)
,
(
(
1
3
)
In Equation
(11),
α
,
β
and
γ
are rep
r
e
s
ent the
wei
g
h
t pa
ramete
rs, SSIM re
sp
ectively
comp
ari
ng from three a
s
p
e
cts: the com
pari
s
on in lu
minan
ce (
l
(x
, y
)), in contr
a
st (
c
(x
, y
)) and
in stru
ctur
e (
s
(x
, y
)).
Takin
g
the 5
1
2
×5
12
standa
rd g
r
ay-level
Lena
as te
sti
ng imag
e an
d
32×32 bin
a
ry
image
as waterm
ark, Figure 2 a
r
e
the simulatio
n
results:
(a) O
r
igin
al image
(b)
Wate
r
ma
r
k
(c
) Wate
r
ma
r
k
ed ima
g
e
(d) E
x
t
ra
ct
watermark
Figure 2. Embeddi
ng and
Extracting W
a
terma
r
k
Shown in Fig
u
re 2
(
a) a
nd (c), the wate
rmarked ima
g
e
has no diffe
ren
c
e with th
e origin
al
image, the
r
ef
ore it
ha
s the
ch
ara
c
te
risti
c
of
goo
d im
perceptivity. At the same
time, there
is n
o
distortio
n
in extract wate
rmark in Figu
re 2(d)
. In experime
n
t, the para
m
eters are set to that
BER=
0
, NC=1.
To refle
c
t the
perfo
rma
n
ce
of improved
thre
shol
d m
odule, thi
s
o
r
iginal meth
od
will be
comp
ared wit
h
pro
p
o
s
ed
method. Met
hod A is the
method in
referen
c
e [8].
Method B is the
algorith
m
that it is the same as metho
d
A but waterm
ark i
s
emb
e
d
ded in the lo
w frequ
en
cy sub
-
band
at th
e f
o
rth l
e
vel. Me
thod
C i
s
th
e
propo
se
d m
e
thod. In
met
hod
D, l
u
min
ance i
s
ch
an
ged
while vari
able
s
are not add
ed. A variable
is adde
d but
the luminan
ce is un
cha
n
g
ed in metho
d
E.
Comp
ari
s
o
n
result
s are
sh
own in Ta
ble
1.
Table 1. Experime
n
tal Re
sults of Meth
od Com
p
a
r
in
g
method
method A
method B
method C
method D
method E
PSNR
35.90
40.15
48,75
48.42
47.22
SSIM
0.988
0.999
0.999
0.999
0.999
T
3.5694
0.3475
0.2927
0.3193
0.3477
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4107 – 41
14
4112
In Tabl
e 1, P
S
NR an
d SSIM are
both
chang
ed
wh
en
ever th
e mo
d
e
l is
chan
ged
or not.
Comp
ari
ng m
e
thod A with method B, it
is obvio
u
s
that when wate
rmark emb
e
d
ded in the low
freque
ncy the
quality of image
s is better than in t
he i
n
terme
d
iate or high fre
q
u
ency. Com
p
a
r
ing
method
s C,
D and E, ch
a
nging lumi
na
nce o
r
varia
b
l
e
s of mod
e
l can incre
a
se
subje
c
tive qua
lity,
but they are
slightly le
ss t
han b
o
th ch
a
nging. Fo
r th
e com
p
licate
d
cal
c
ul
ation
in equ
ation (3) of
origin
al meth
od, the time is far mo
re th
an that of the referen
c
e mo
del.
3.2. Compari
s
on of Alg
o
r
i
thms
For refle
c
ting
the universality of algorithm, waterm
ark i
s
embe
dded in five different
types of imag
es. Table 2
compa
r
ed al
go
rithm with ref
e
ren
c
e [8] an
d referen
c
e [10].
Table 2. Experime
n
tal Re
sults of Diffe
rent Images
The tested imag
e
Fang’s
method [10]
Bar
n
i’s
method [8]
Barni’s method and
embedding in LL
3
Proposed metho
d
PSNR
milkdrop 40.23
34.49
40.18
42.58
man
36.23
30.85
38.10
43.32
couple 36.23
33.62
38.82
44.42
plane 36.98
33.69
37.99
44.42
lake 33.85
31.57
36.76
42.58
SSIM
milkdrop 0.988
0.994
0.999
0.999
man 0.994
0.996
0.999
0.999
couple 0.994
0.995
0.999
0.999
plane 0.996
0.992
0.998
0.999
lake
0.992
0.993
0.998
0.999
In Table
2, five kind
s eval
uation of im
a
ge qu
ality ap
plied o
n
the
prop
osed m
e
thod a
r
e
sup
e
rio
r
to the others. The
r
efore, Table
2
confirmed t
he feasi
b
ility of propo
se
d method.
For d
e
scri
bin
g
the ro
bu
stness of the
prop
osed al
g
o
rithm, Meth
od 1 is th
e method in
referen
c
e [1
0
], method
2 i
s
the
Huan
g' s
metho
d
[1
1]. Method
1
and
metho
d
2 a
r
e
comp
ared
with the prop
ose
d
algo
rith
m whi
c
h re
prese
n
ts the m
e
thod 3 in th
is pap
er. Cro
pping, ad
din
g
noise, filterin
g, scalin
g an
d rotatio
n
in
three
kin
d
s o
f
method
s
re
spe
c
tively, the expe
riment
al
results a
r
e sh
own in Ta
ble
3.
Table 3. Simple Attack
Attack
Cropping
Scaling
Adding
noise
Filtering
Rotation
1/8 1/2
25%
200%
Gaussian
0.005
Speckle
0.05
Median
5×5
Lo
w
and
pass
5° 30°
PSNR
1
15.37
8.46 29.85
38.77
22.92
18.25
32.86
38.54
19.42
13.58
2
15.49
8.46 29.98
42.97
22.99
18.30
32.17
41.84
15.24
13.59
3
18.19
8.47 29.87
42.68
32.69
33.62
33.99
40.73
19.45
13.60
SSIM
1 0.851
0.499
0.667
0.952
0.523
0.520
0.701
0.957
0.938
0.783
2 0.874
0.499
0.670
0.955
0.515
0.521
0.705
0.955
0.349
0.782
3 0.875
0.501
0.669
0.956
0.673
0.735
0.703
0.960
0.942
0.788
BER
1 0.03
0.28
0.15
0
0.02
0.06
0.09
0.01
0.01
0.07
2 0.05
0.28
0.35
0.08
0.23
0.31
0.27
0.09
0.05
0.16
3 0.04
0.24
0.08
0
0.06
0.02
0.17
0
0.001
0.06
NC
1
0.977
0.810
0.901
1
0.512
0.960
0.940
0.999
0.992
0.957
2 0.954
0.811
0.772
0.949
0.855
0.801
0.821
0.938
0.967
0.904
3
0.967
0.864
0.952
1
0.670
0.986
0.900
1
0.993
0.956
From th
e PSNR
and SSIM
sho
w
n i
n
Ta
ble 3, metho
d
2 still have
hi
gher
num
eri
c
al value
after the wat
e
rma
r
ked attacked in whi
c
h sho
w
ed g
ood impe
rce
p
tivity.However,it can be
seen
from the
BER and
NC that
the extra
c
ted
wate
rma
r
k o
f
method
2 i
s
wo
rst.
Com
p
aring
meth
od
1
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Adaptive
All-odd T
r
an
sform
a
tion Wa
term
ark Sche
m
e (Qiaom
ei Ma)
4113
and metho
d
3 in the four sets of data, the forme
r
attacked
watermarked ima
g
e
s an
d analy
s
is o
f
watermark e
x
tracted a
r
e
both slig
htly worse
th
an the latter. Th
us, propo
se
d
method resi
sts
these five kin
d
s of attacks i
n
a cert
ain scope.
The JPEG
comp
re
ssion
is implem
e
n
ted to the
waterm
arke
d image
s, and the
experim
ental
results a
r
e prese
n
ted in Fi
gure 3.
(a)
(b)
Figure 3. JPEG Com
p
re
ssi
o
n
In the Figure
3(a
)
, it is obvious that PSNR of
three me
thods in
crea
se with the increa
sing
of the quality
factor
(QF
)
.
PSNR of m
e
thod 3 i
s
bi
gg
er than
that
of method
1
but sm
aller t
han
method 2. Th
erefo
r
e, impe
rce
p
tivity of
waterm
arke
d
image
s is th
e best in m
e
thod 2. From
the
Figure 3
(
b
)
, robu
stne
ss of
method
3 i
s
t
he
stro
nge
st
and
extra
c
ted
wate
rma
r
k i
s
light
disto
r
tio
n
whe
n
QF=10.
Howeve
r, it i
s
disto
r
tion when QF
=10
0
and co
mplete
ly distorted when QF
=20 i
n
method 2.
From t
he a
n
a
lysis of the
wate
rma
r
ki
n
g
mod
e
l, p
r
o
posed m
odel
gre
a
tly enh
ance the
robu
stne
ss a
nd imp
e
rcepti
v
ity of watermarking
sc
he
me, the valu
e
of pe
rform
a
n
c
e
evaluation
is
better than ot
her meth
od
s, and alg
o
rithm
model ha
s b
een imp
r
oved
greatly.
4. Conclusio
n
Analyzing the
feature of the watermarki
ng sc
hem
e in
this pape
r, the ne
w wate
rmarking
model, whi
c
h
improve
s
thresh
old calcul
ation
method
base
d
on th
e odd
-even
quanti
z
ation,
is
prop
osed
on
the ba
si
s of
the Ba
rni’
s m
odel. T
he
m
o
del e
n
sure
s i
m
perce
ptivity of the
watermark
by calculatin
g the JND a
nd reg
u
lating
thre
shold
with improved
method, and
enhan
ce
s th
e
robu
stne
ss
with the help
o
f
scrambli
ng
and o
dd-ev
e
n
qua
ntizatio
n of the
wate
rmark. Mo
re
o
v
er,
blind extracti
on
of wate
rm
ark can
be
realized by
u
s
ing all
-
od
d transfo
rmatio
n. Co
mpa
r
ing
with
the existing
a
l
gorithm
s, the
pro
p
o
s
ed
m
odel h
a
s ma
de g
r
eat p
r
o
g
r
ess i
n
the
a
b
ility of resi
st
ing
geomet
ric
attacks. T
h
is
m
odel i
s
not
co
nfined to
cov
e
r, the
sele
cti
on of waterm
ark emb
edd
e
d
is
also
not ai
me
d to the
gray image
or a
binary i
m
age,
col
o
r i
m
age
s, text files, video, a
udio, e
t
c.
can al
so a
c
t the both. The
appli
c
ation of
the
waterm
arking
scheme
will be a go
od
extension.
Ackn
o
w
l
e
dg
ements
This
work
wa
s suppo
rted
by Natural Scien
c
e F
oun
d
a
tion of No
rt
h Unive
r
sity of China
and scie
ntific and technol
o
g
ical
key-p
r
oj
ect in
Shanxi Province of China (No.20
0
9032
2004
)
Referen
ces
[1]
Xi
nsh
an Z
hu,
Jie Di
ng. A Novel Qua
n
tizat
i
on W
a
termark
ing Sch
e
me U
s
ing R
and
om
Normal
i
ze
d
Correl
a
tio
n
Mo
dul
ation.
C
h
in
e
s
e Journ
a
l of C
o
mputers
. 20
1
2
; 35(9): 19
59-
197
0.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 5, May 2014: 4107 – 41
14
4114
[2]
Lili Zh
ou, Gu
a
ng Wu, Ya
n H
e
, et al.
A n
e
w
Just-Notice
abl
e-Distortio
n
mo
del c
o
mbi
n
e
d
w
i
th the de
pth
infor
m
ati
on a
n
d
its app
licati
o
n in Multi-v
i
ew
Video C
o
d
i
ng
.
Proceed
in
gs of the 201
2 8t
h Internati
ona
l
Confer
ence
on
Intelli
gent Inf
o
rmatio
n
Hi
din
g
an
d Multim
e
d
ia S
i
gn
al Pr
o
c
essin
g
. Gree
ce. 201
2; 66:
246-
251.
[3]
A Beg
h
d
adi,
MC Lar
ab
i, A
Bouzer
do
um, et al. A
surve
y
of p
e
rcept
ua
l ima
g
e
proc
e
ssing
metho
d
s
.
Sign
al Process
i
ng: Image C
o
mmu
n
icati
o
n
. 2
013; 28(
8): 811
-831.
[4]
Bhag
ya
la
xm
i S
Kull
oll
i
, Vij
a
y
S Raj
pur
ohit,
Anil B
Gavad
e
.
An Alg
o
rithmi
c Appro
a
ch f
o
r
Multisp
e
ctral
Image Qua
lit
y
Assessment.
Internati
o
n
a
l Jo
urna
l of Curr
e
n
t Engi
ne
erin
g
and T
e
c
h
n
o
lo
gy
. 201
3; 9:
209-
214.
[5]
Sung-
Ho Ba
e, Munchur
l Kim
.
A Novel DC
T
-
Bas
ed JND Mode
l for Lum
ina
n
ce Ad
apta
t
ion Effect in
DCT
F
r
equenc
y.
IEEE Signal Processing Let
ters
. 2013; 20(
9): 893-8
96.
[6]
Hon
gbo Bi,
Xuemi
ng Li, Yu
bo Z
han
g. A Novel HV
S-ba
sed W
a
termar
k
ing Sch
e
me i
n
Conto
u
rlet
T
r
ansform Domain.
T
E
LKOMNIKA Indone
sian Jo
urna
l of Electrical En
gi
neer
ing
. 20
13;
11(12): 75
16-
752
4.
[7]
Humin
g
Ga
o,
Li
yu
an J
i
a, M
e
ilin
g
Liu.
A D
i
g
ital
Watermar
k
ing
Alg
o
rithm
for C
o
lor
Ima
ge B
a
se
d o
n
DWT.
T
E
LKOMNIKA Indone
sian Jo
urna
l of Electrical E
ngi
neer
ing
. 2
013;
11(6): 32
71-
32
78.
[8]
Barni M, Bartol
ini F
,
Piva A. Improve
d
w
a
v
e
l
e
t-base
d
w
a
ter
m
arking thr
o
u
g
h
pi
xel-
w
i
s
e
m
a
skin
g
.
IEEE
T
r
ansactio
n
s o
n
Ima
ge Proce
ssing
. 20
01; 5(
10): 783-
79
1.
[9]
Hon
g
Sh
en,
Bo C
hen.
F
r
o
m
sin
g
le
w
a
te
rmark to
dua
l
w
a
term
ark:
A ne
w
a
ppr
o
a
ch for
ima
g
e
w
a
t
e
rmarking.
Co
mp
uters an
d Electrica
l
En
gin
eeri
n
g
. 20
1
2
; 38(5): 13
10-
132
4.
[10]
W
angsh
e
n
g
F
ang, Kan
g
Ch
en.
A w
a
velet w
a
termarki
ng
base
d
on HVS
and w
a
termar
king ca
pacity
analysis
. 1st Internati
o
n
a
l C
onfere
n
ce o
n
Multimed
ia Inf
o
rmatio
n
Net
w
orkin
g
and Se
curit
y
. Hu
be
i.
200
9; 2: 141-1
44.
[11]
Ersong H
u
a
n
g
,
Jinhu
a Li
u, Ruh
ong W
e
n. Mult
ipl
i
cativ
e
w
a
t
e
rmarki
ng
alg
o
rithm b
a
se
d on
w
a
vel
e
t
visual m
ode
l.
Journ
a
l of Co
mputer App
lic
ations
. 201
1; 31(
8): 2165-
21
73.
Evaluation Warning : The document was created with Spire.PDF for Python.