TELKOM
NIKA
, Vol. 11, No. 12, Decem
ber 20
13, pp.
7516
~75
2
4
e-ISSN: 2087
-278X
7516
Re
cei
v
ed
Jun
e
27, 2013; Revi
sed Aug
u
st
11, 2013; Accepted Sept
em
ber 2, 201
3
A Novel HVS-based Watermarking Scheme in
Contourlet Transform Domain
Hongb
o Bi*
1,2
, Xueming L
i
1
, Yubo Zhang
2
1
Beijin
g Ke
y L
a
borator
y of Net
w
o
r
k S
y
stem
a
nd Net
w
o
r
k Cu
lture, Beij
ing, C
h
in
a
2
School of Elec
trical Informati
on Eng
i
n
eeri
n
g
,
NorthEast Petrole
u
m Univ
ers
i
t
y
, Da
qin
g
, Chi
n
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: bhbd
q@1
26.
com
A
b
st
r
a
ct
In this pa
per, a
nove
l
w
a
termarkin
g techn
i
q
u
e
in co
ntour
let
transform (
C
T
)
do
mai
n
is
pre
s
ented.
T
he pro
pose
d
alg
o
rith
m take
s adva
n
tage
of a mu
ltisca
le
framew
ork an
d mu
lti-dir
e
ctio
na
lity to extract th
e
signific
ant freq
uency, l
u
min
a
n
c
e an
d texture
compo
nent
i
n
an i
m
a
ge. U
n
li
ke the co
nventi
ona
l metho
d
s i
n
the conto
u
rlet
do
ma
in, mask function
is acc
o
mplis
he
d pixe
l by pixe
l by taking i
n
to acco
u
n
t the freque
nc
y,
the lu
mi
na
nce
and th
e textur
e cont
e
n
t of al
l the i
m
a
ge su
bba
nds i
n
clu
d
i
ng the l
o
w
-
pas
s subba
nd
an
d
directi
ona
l sub
ban
ds. T
he ad
aptive n
a
ture
o
f
the novel
met
hod a
l
l
o
w
s
the sche
m
e to b
e
ada
ptive i
n
terms
of the imp
e
rce
p
tibil
i
ty and ro
bustness. T
he
w
a
termar
k is d
e
tected by co
mp
utin
g the co
rrelatio
n
. F
i
nal
l
y
,
the ex
peri
m
e
n
tal res
u
lts d
e
monstrate t
h
e i
m
p
e
rc
epti
b
i
lity an
d the
robustn
ess a
g
a
inst sta
ndar
d
waterm
arking attacks.
Ke
y
w
ords
:
digital watermarking, hum
a
n visual
system
,
m
a
s
k
function, blind detection
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
The ra
pid de
velopment of multimedia te
c
hni
que a
nd
comp
uter n
e
tworks broug
h
t
about
the issue of copyri
ght pro
t
ection.
Digit
a
l watermarki
ng ca
n be co
nsid
ere
d
as
a solutio
n
to the
probl
em of i
n
tellectu
a
l p
r
operty
rights (IPR)
of dat
a content
s, whi
c
h e
m
be
d
ded the
secret
identity information into the host data i
n
su
ch a
way
that it usuall
y
exhibits imperceptibility and
robu
stne
ss a
gain
s
t the intentional o
r
u
n
intentio
n
a
l o
peratio
ns
su
ch as
com
p
re
ssion, line
a
r a
n
d
nonlin
ear filtering, n
o
ise a
nd g
eomet
ric tran
sfor
mations [1-3]. Digital wa
te
r
m
ar
k
i
ng
ha
s b
e
e
n
widely ap
plie
d in many fields such as
content aut
he
n
t
ication, finge
rpri
nting, and
servi
c
e tra
c
i
ng
etc
[4, 5].
Imperceptibili
ty and ro
bu
stne
ss
are two m
o
st
im
portant
req
u
i
r
eme
n
ts of
digital
watermarkin
g
.
For meet these two requi
reme
nts sim
u
ltaneou
sly, people have p
r
opo
se
d man
y
scheme
s
in
either
spatial
or tran
sform
domain.
Sp
ace d
o
main
method
s hid
e
informatio
n
by
cha
ngin
g
sp
ace
dom
ain
cha
r
a
c
teri
stic of ho
st d
a
ta, whil
e tra
n
s
form
do
mai
n
metho
d
s b
y
changing some coeffici
ents in transform
dom
ain of host data. The familiar watermark
algorith
m
s in
transfo
rm d
o
main in
clud
e discrete fo
urie
r tran
sform (DF
T
), di
screte
co
sin
e
transfo
rm
(D
CT), di
s
c
rete
wav
e
let tra
n
sform (DWT), sin
gula
r
value de
com
positio
n (SV
D
),
discrete
cont
ourlet tra
n
sfo
r
m (CT), an
d so on [6
-8].
As an efficie
n
t geometri
c repre
s
e
n
tation of
natural
images, CT
has attra
c
te
d many
resea
r
chers’
eyes. CT i
s
a true
re
presentation of the digital imag
e, which provides a flexibl
e
multire
s
olutio
n representa
t
ion for two
-
dimen
s
ion
a
l
sign
al [9].
Comp
ared
with DWT,
CT
posse
sse
s
the cha
r
a
c
teri
stics of the di
rectio
nality and ani
sotro
p
y
. Since its coeffici
ents a
r
e
spa
r
se, an
d
repre
s
e
n
t the
intrinsi
c
prop
erty, CT
ha
s
been
wi
dely
use
d
in
a va
riety of image
pro
c
e
ssi
ng
such
as
com
p
re
ssi
on, de
noisi
ng, enh
ancement, e
t
c. Mean
whil
e, resea
r
che
r
s
bega
n to appl
y CT to the image waterm
arki
ng [10
-
13]
.
This pa
per p
r
opo
se
s an
adaptive wat
e
rma
r
ki
ng m
e
thod in co
n
t
ourlet domai
n. The
mask of fre
quen
cy, luminan
ce an
d texture by h
u
man eye
s
has b
een e
x
amined in t
he
frame
w
ork of conto
u
rlet tra
n
sfor
m tech
ni
que. Accordi
ng to the results, the emb
eddin
g
stre
ng
th
of wate
rma
r
k com
pon
ents is
determin
ed a
daptiv
el
y. Experimen
tal re
sults d
e
mon
s
trate
the
novel embe
d
d
ing strategy
and demo
n
s
trate that
th
e prop
osed
watermarkin
g
algorithm a
r
e
invisible and
very
ro
bu
st
a
gain
s
t
noi
se and co
m
m
on
image
proce
ssi
ng te
ch
niq
ues such a
s
lossy comp
re
ssi
on, filtering
,
croppi
ng an
d noise additi
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel
HVS-ba
sed
Wate
rm
arking Sche
m
e
in Contou
rlet Tra
n
sfo
r
m
Dom
a
in (Hongb
o BI)
7517
The o
u
tline
of the rest
of th
e pa
per is organi
zed
as fo
llows. In
se
ction 2, th
e con
t
ourlet
transfo
rm al
gorithm is m
entione
d, the Laplac
i
an p
y
ramid and
dire
ctional fil
t
er ban
ks a
r
e
introdu
ce
d. In se
ction 3,
we de
scri
be
the ju
st noti
c
eable di
sto
r
tion profile of human vi
sion
,
whi
c
h
co
nsi
d
ers the
influe
nce
of freque
ncy, lumin
a
n
c
e
and
texture. In sectio
n
4, the p
r
o
posed
watermarkin
g
scheme i
s
pre
s
ente
d
, we give
the e
m
beddi
ng an
d the blind d
e
tection of th
e
watermarkin
g
.
In sectio
n 5,
some
experi
m
ental
results are mentio
ned. Finally,
con
c
lu
sio
n
s a
r
e
given.
2. The Con
t
o
u
rlet Tran
sfo
r
m
D
W
T
pr
o
v
id
es
th
e
c
h
ar
acte
r
i
s
t
ics o
f
mu
lti-
sc
a
l
e d
e
co
mp
os
itio
n an
d
time-
f
r
equ
e
n
c
y
locali
zation
whi
c
h facilities th
e representation
of the image.
However it
can only
offer
the
informatio
n in few directi
ons a
nd lo
w-pass co
mp
on
ents. The
co
ntourlet tra
n
sform is a
no
vel
image
de
com
positio
n
sche
me, whi
c
h
ca
n capture th
e
intrin
sic ge
o
m
etrical
stru
cture in
visu
al
informatio
n. It exploits the
Lapla
c
ian
Pyramid (LP)
fo
r the multire
s
o
l
ution de
com
positio
n of th
e
image. Afterwards, a dire
ctional de
co
mposit
io
n is
perfo
rmed o
n
every bandp
ass image u
s
ing
dire
ctional filt
er b
a
n
ks. It c
an obtai
n a
sparse
expan
sion for
nat
ural image
by
spe
c
ifying th
e
numbe
r of di
rectio
nal b
a
n
d
s at a
n
y level. As
a resu
lt, the image
is re
present
ed a
s
a
set
of
dire
ctional
su
bban
ds at mu
ltiple scale
s
.
CT is illustrat
ed in Figure 1.
I
m
ag
e
↓
(M,
M
)
DF
B
DF
B
LP
LP
Figure 1. Block
Diag
ram o
f
CT
Comp
osed of
low-pa
ss filtering
and d
o
w
n
s
ampl
i
ng,
LP is inde
ed
the high fre
quen
cy
comp
one
nts
of the gaussi
an pyrami
d
(GP) at the sa
me scale, tha
t
is, the detailed pa
rts of the
image. LP i
m
age
can
b
e
obtaine
d
by subtra
ctin
g the two n
e
ighb
ouri
ng i
m
age
s in G
P
.
Gene
rally, we nee
d to expand th
e fine
r scale i
m
age
to the co
arser scal
e, nam
ely, perform t
h
e
interpol
ation t
o
the
rows
an
d col
u
mn
s of
the im
ag
e, afterwards, the
i
n
terpol
ated
result
s throug
h
a low-pass filter will
subt
ract t
he image
at the co
arser scale. T
he
reconstr
uction of LP is the
inverse of the decom
po
sition.
Suppo
se th
e image
)
(
2
2
L
IMG
, L
P
in CT
use
s
o
r
thog
onal filters and
downsamplin
g by 2 in each dime
nsi
o
n
,
the
l
-level o
f
LP decomp
o
se
s
IMG
into a
coarser
image
J
V
and a seq
uen
ce of detail image
j
I
, where
J
V
is the approximatio
n sub
s
p
a
ce a
t
the
scale
l
(
J
l
2
).
I
can
be den
oted a
s
:
)
(
)
(
1
2
2
j
J
j
J
I
V
L
(1)
The di
re
ction
a
l filter ban
k
(DFB
) is
gen
erally impl
em
ented via a
n
l
-level bin
a
ry
tree
decompo
sitio
n
that le
ad
s t
o
l
2
sub
ban
ds with wed
ge-shape
d
fr
e
que
ncy
p
a
rtition as sh
own
in
Figure 2. In
CT, the simpl
i
fied DFB is intuitivel
y constru
c
ted from
two bu
ilding
blocks. The first
building bl
ock is a two-cha
nnel quin
c
u
n
x
filter bank
with fan filters that
divides a 2-D sp
ect
r
um
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 751
6 – 7524
7518
into two directions: ho
rizon
t
al and vertical. The se
co
n
d
buildin
g blo
ck of the
DF
B is a sh
eari
n
g
operator, whi
c
h amo
unts t
o
just reo
r
d
e
ri
ng of image sample
s.
0
y
1
y
0
H
1
H
1
2
l
H
0
S
1
S
1
2
l
S
0
S
1
S
1
2
l
S
0
G
1
G
1
2
l
G
x
1
2
l
y
x
Figure 2.
l
-level Binary Tre
e
De
comp
osit
ion
In CT, the DFB partition further
j
I
from LP. The results a
r
e detail su
bb
and
s in multiple
dire
ction
s
.
2
,
2
0
)
,
(
,
1
Z
k
j
I
I
j
j
l
l
k
j
k
j
(2)
Figure 3 sh
o
w
s a
n
exampl
e of CT on th
e “Len
a” ima
ge.
Figure 3. De
compo
s
ition of
Lena by CT
The im
age
i
s
decompo
se
d
into a
lo
w-p
a
s
s subb
and
a
nd a
set of
di
rectio
nal
sub
band
s.
We
notice th
at CT
effe
ctively rep
r
e
s
e
n
ts the
tru
e
i
m
age
where
edge
s
are lo
cali
zed
in
bo
th
locatio
n
and
dire
ction.
3. Just Notic
eable Dis
t
or
tion
(JND) Profile Anal
y
s
i
s
in CT
The co
ntou
rl
et transfo
rm provide
s
a m
u
ltis
cale and
multidire
c
tion
al rep
r
e
s
enta
t
ion of
an imag
e. It is ea
sily adj
u
s
table fo
r det
ecting fin
e
te
xture detail i
n
any ori
enta
t
ion at variou
s
scale level. I
n
order to
conform t
o
th
e mult
ire
s
ol
u
t
ion natu
r
e o
f
human
visu
al syst
em an
d
enha
nce the
performan
ce of the waterma
r
k
sy
ste
m
, we cal
c
ul
ate the weig
ht param
eters
according to the frequ
en
cy, luminan
ce a
nd texture co
mplicatio
n of detail su
bba
n
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel
HVS-ba
sed
Wate
rm
arking Sche
m
e
in Contou
rlet Tra
n
sfo
r
m
Dom
a
in (Hongb
o BI)
7519
In order to e
m
bed into the host the maximu
m, but still impercep
t
ible, HVS has to be
con
s
id
ere
d
. Lewi
s
and K
nowl
e
s tackl
e
the proble
m
of DWT coeffici
ents
quanti
z
ation
for
comp
re
ssion
purpo
se
s.
Detail
s of t
he imp
r
oved
HVS-b
a
sed
wate
rma
r
ki
ng meth
od
are
pre
s
ente
d
by Barni
et
al. i
n
20
01
and
some m
odifica
tions
of the
model
is prop
ose
d
in
o
r
de
r to
better fit the
model
to th
e waterm
arki
ng
system,
whi
c
h
has b
een
widely
cited by ma
n
y
resea
r
cher.
Ho
wever,
DWT can o
n
ly provide info
rmation in
3
dire
ction
s
, that is, hori
z
o
n
t
al,
vertical
an
d d
i
agon
al, while
CT
can
de
compo
s
e
the
host
alon
g a
r
bitrary
directi
ons at
arbit
r
a
r
y
scale, which
provid
es m
o
re
details i
n
form
atio
n th
an DWT
and
can
be
co
n
s
ide
r
ed
as the
potential sol
u
tion to improve the perform
ance of
the waterma
r
ki
ng system. Detail
s of the HVS
cha
r
a
c
teri
stics in CT a
r
e m
entione
d here
.
The
Ju
st
N
o
t
i
cea
b
le
Dist
o
r
t
i
on (
J
N
D
)
in
t
he CT
dom
a
i
n can b
e
typi
cally exp
r
e
s
sed a
s
the produ
ct
of three
terms. As a
re
sult, the
HV
S mask fu
n
c
tion
i
s
ada
ptive to the
CT
coeffici
ents.
Con
s
id
erin
g
the se
nsitivity of the human
eye, Xiao et al. propo
sed the
weig
ht
coeffici
ent cal
c
ulatio
n as
sh
own b
e
lo
w:
25
.
0
)
,
,
(
)
,
,
(
)
,
(
)
,
,
(
y
x
J
M
y
x
J
M
J
k
M
y
x
J
M
t
l
f
(3)
Whe
r
e
)
,
(
J
k
M
f
de
notes frequ
ency
sen
s
itivity as sho
w
n in Eq
u
a
tion (4
).
)
,
,
(
y
x
J
M
l
denote
s
lo
ca
l luminan
ce
for g
r
ay level
s
in with
refe
re
nce to E
quati
on (5) a
nd
(6
).
)
,
,
(
y
x
J
M
t
denote
s
the influen
ce of the texture as i
ndicated in th
e Equation (7
), (8) a
nd (9
).
4
,
1
.
0
3
,
16
.
0
2
,
32
.
0
1
,
1
3
4
,
2
2
4
,
2
1
4
,
1
4
,
2
2
)
,
(
J
J
J
J
n
k
n
k
n
k
n
k
J
k
M
f
(4)
)
,
,
(
)
(
)
,
,
(
)
,
,
(
)
(
)
,
,
(
)
(
)
(
)
,
,
(
y
x
J
L
J
L
y
x
J
L
y
x
J
L
J
L
y
x
J
L
J
L
J
L
y
x
J
M
l
(5)
1
44
1
,,
,
25
6
2
2
k
JJ
xy
LJ
x
y
c
(6)
)
,
,
(
)
,
,
(
)
,
,
(
y
x
J
D
y
x
J
E
y
x
J
M
t
(7)
2
41
1
1
44
00
0
,,
,
22
J
k
JJ
ij
xy
EJ
x
y
c
j
i
(8)
1
44
0,
1
0,
1
,,
1
,
1
22
k
JJ
i
j
xy
DJ
x
y
c
j
i
(9)
Whe
r
e
,
k
cx
y
deno
tes the d
e
co
mpositio
n co
efficient at
)
,
(
y
x
in the subb
an
d
k
.
,,
E
Jx
y
denote
s
the local
sum of squares,
,,
D
Jx
y
deno
tes the local varian
ce.
Assu
ming
ch
ange
s
smalle
r than o
ne h
a
lf of t
he cal
c
ulate
d
ma
sk function
are
visuall
y
imperceptible,
)
,
,
(
y
x
J
M
gives max
i
mum emb
e
d
d
ing threshol
d in the q
u
a
n
tization
of CT
coeffici
ents u
s
ing:
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 751
6 – 7524
7520
2
/
)
,
,
(
)
,
,
(
y
x
J
M
j
i
J
T
(10)
Acco
rdi
ng to
the mentio
n
above, it is appa
rent th
at the comp
uted ma
sk functio
n
)
,
,
(
y
x
J
M
at each
pi
xel enable
s
HVS-ba
se
d
watermarki
ng to obtai
n high level
of
imperce
ptibility and robu
stn
e
ss witho
u
t consi
der
ation
of the comput
ations in e
q
u
a
tions.
4. Proposed
Wa
termar
king Sy
stem
4.1. Watermarking Embe
dding
CT coefficie
n
ts contain
an app
roxi
mate low-pa
ss
su
bban
d
and several detail
dire
ctional
su
bban
ds at ea
ch level. The low-pa
ss
sub
band represe
n
ts ba
sic info
rmation of the
image, which
is the mo
st importa
nt part
for im
age re
con
s
tru
c
tion, embed
ding
th
e
wate
rma
r
k
i
n
these
coeffici
ents m
a
y a
c
h
i
eve ro
bu
stne
ss ag
ainst
i
n
tentional or un
intentional attacks, but
ma
y
degrade th
e visual qu
ality, since th
e h
u
man vis
ual
system
(HVS
) is le
ss
se
n
s
itive to high
freque
nci
e
s,
embeddi
ng
the waterm
ark in t
he
high freq
uen
cy sub
ban
ds improve
s
the
perceptibility of the waterm
arked im
a
ge, but it is hardl
y robust.
In our sch
e
m
e
, as shown i
n
Figure 4, t
he wate
rma
r
k is em
bed
d
ed into both t
he low-
pass subb
an
d and the direction
a
l su
bb
and
s by di
fferent HVS ch
ara
c
teri
stics. Con
s
e
quently
,
the pro
p
o
s
ed
watermarkin
g
schem
e is robu
st to
the widely spe
c
tral attacks
re
sulting from b
o
t
h
the low an
d h
i
gh frequ
en
cy image proce
ssi
ng.
Figure 4. W
a
t
e
rma
r
ki
ng Scheme
The ge
ne
ral
embed
ding
st
eps fo
r the p
r
opo
se
d wate
rmarkin
g
sch
e
me a
r
e d
e
scrib
e
d
as
follows
.
Step 1. Watermarking p
r
ep
rocessin
g
The wate
rma
r
k info
rmation
need be p
r
e
p
ro
ce
ssed
in
orde
r to wea
k
en the co
rrela
t
ion of
watermark i
m
age pixel
s
and enh
an
ce sy
stem robu
stne
ss. In our sch
e
me, we trea
t the
watermark im
age u
s
ing Arnold scrambl
e
as sho
w
n in
Equation (1
1
)
.
)
(mod
2
1
1
1
N
y
x
y
x
(11)
Whe
r
e
)
,
(
y
x
is the pixel of the wate
rma
r
ki
ng image,
)
,
(
y
x
is the pixel o
f
the
watermarkin
g
image after scra
mble.
Since th
e Arnold t
r
an
sfo
r
m is
pe
riod
ic, the nu
m
ber
of scra
mbling
can
be al
so
con
s
id
ere
d
a
s
the key to e
nhan
ce the
secu
rity.
Step 2. CT of the whole im
age
We
pro
p
o
s
e
to emb
ed t
he
watermark in
the lo
w-pa
ss subb
a
nd a
nd o
n
e
of the
dire
ctional
su
bban
ds of th
e hig
h
e
s
t lev
e
l, that i
s
, J
th
level,
which
actu
ally a
r
e
mid-fre
que
ncy
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel
HVS-ba
sed
Wate
rm
arking Sche
m
e
in Contou
rlet Tra
n
sfo
r
m
Dom
a
in (Hongb
o BI)
7521
sub
ban
ds i
n
t
e
rm
s of the
whol
e fre
que
ncy ba
nd
of
the ima
ge. Th
is is an
app
ropriate
trad
e
o
ff
betwe
en the
visual pe
rcep
tibility and the robu
stne
ss.
Step 3. Determination of the embed
ding
positio
n
Since
HVS is le
ss
sen
s
iti
v
e to the texture, the waterma
r
ki
ng e
m
beddi
ng di
rection
a
l
sub
ban
d can
be cho
s
en
combi
ned
with the text
ure in the image. The energy distributi
o
n
corre
s
p
ond
s
to the texture pro
perty, so we em
bed
the waterm
arki
ng in the
most textured
dire
ctional su
bban
d.
The ene
rgy is compute
d
as Equation (1
2
)
.
d
Jd
J
M
x
M
y
d
J
d
J
d
J
d
J
y
x
C
M
M
E
,,
11
2
,
,
,
,
)
,
(
1
(12)
W
h
er
e
d
J
M
,
rep
r
e
s
ent
s the
wi
d
t
h and
hei
ght
of the
d
th
di
rection
a
l
sub
b
and i
n
the
J
th
level,
d
J
C
,
represents the
corresp
ondi
ng co
efficients,
)
,
(
y
x
is the co
ordinate
in the sub
b
a
nd.
The bi
gge
r
d
J
E
,
is, the m
o
re
texture i
s
, whi
c
h m
ean
s th
e su
bba
nd
contribute
s
m
o
re to
th
e
image.
Step 4. Embedding the
wat
e
rma
r
k
Acco
rdi
ng to the freque
ncy
,
luminance a
nd te
xture co
mplicatio
n value of the sub
band
,
a bina
ry wa
termark i
s
e
m
bedd
ed int
o
the ho
st i
m
age, by m
odifying the
value of th
e
corre
s
p
ondin
g
coeffici
ents.
Thus, the em
beddi
ng can
be de
scribe
d as:
)
,
(
)
,
,
(
)
,
(
)
,
(
j
i
W
y
x
J
T
j
i
IM
j
i
IM
w
(13)
Whe
r
e
)
,
(
j
i
IM
w
re
p
r
esents the
wate
rma
r
ke
d subba
nd
coeffici
ents,
)
,
(
j
i
IM
rep
r
e
s
ent
s the origin
al su
bban
d co
efficient
s,
deno
tes the glo
b
a
l embe
ddin
g
factor that
determi
ne
s the emb
eddi
n
g
strength,
)
,
,
(
y
x
J
T
is the
HVS m
a
sk fun
c
tion
as fo
rmulate
d
in
(10
)
,
)
,
(
j
i
W
denote
s
the bina
ry pse
udo
ran
d
o
m
distri
b
u
ted
waterm
ark, whi
c
h is a
rra
nged in
the form of two dimen
s
ion
s
.
Step 5. Image Re
con
s
tru
c
tion
Finally the waterma
r
ked i
m
age i
s
obtai
ned by
the co
ntourlet re
con
s
tru
c
tion
of subba
n
d
coeffici
ents
combine
d
with
the remain
ed
coefficie
n
ts followed by Arnold scrambl
e
.
4.2. Watermarking Dete
c
t
ion
Maximum-li
kelihoo
d dete
c
tion is u
s
e
d
to extract each emb
edde
d bit from the
watermarke
d
signal
coef
ficients. Th
e
waterm
ar
k is re
covere
d by mean
s of contou
rl
e
t
recon
s
tru
c
tio
n
and calcula
t
ions in reve
rse orde
r of embed p
r
o
c
e
s
s.
Step 1. Perform contou
rl
et transfo
rm
for
the wate
rmarked ima
g
e
to get con
t
ourlet
coeffici
ent of all the sub
b
a
nds a
nd dete
r
mine the su
b
band
with ma
ximum energ
y
.
Step 2. Detect the correlati
on.
The corre
l
ation is calcu
l
ated as:
'
,
11
1
,,
MN
lk
xy
cx
y
w
x
y
MN
(14)
Due to th
ere
are
som
e
di
stortions of
de
tected
wate
rmark in
som
e
deg
ree,
is
set to
be threshold,
if
, the
wat
e
rmark is exis
t, if
P
, the
watermark i
s
not exi
s
t. T
he
threshold
is related to the false al
arm p
r
obab
ility and
false di
smissal prob
ability.
We m
a
ke th
e assu
mption
that
,
,
lk
cx
y
are ze
ro me
an, in
d
epen
dent va
riable
s
. By
exploiting the
central limit theore
m
, we
can
also co
n
s
ide
r
that
is norm
a
lly distributed [14
-
15]. Unde
r th
ese hyp
o
the
s
es, it can
be
easily de
du
ced that the m
ean value
s
of
in ca
se
s A
(not waterm
a
r
ke
d), B (wat
erma
rked
with anoth
e
r wrong waterma
r
k)
an
d
C (waterma
rked with
the c
o
rrec
t watermark) are:
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 751
6 – 7524
7522
:0
A
A
(15)
:0
B
B
(16)
,
11
,,
:,
MN
Cl
k
xy
lk
lk
CE
H
x
y
MN
(17)
Whe
r
e
E
denotes the expe
ct
ation.
The false al
arm probability
Pr
f
Po
b
A
o
r
B
，
for
case A:
2
2
2
,
2
11
,,
,
MN
W
Al
k
xy
lk
l
k
E
cx
y
MN
(18)
For
case B:
2
22
22
2
,,
2
11
,,
,,
MN
W
Bl
k
W
l
k
xy
lk
lk
E
cx
y
E
H
x
y
MN
(19)
Whe
r
e
2
denotes the varia
n
c
e, and:
22
2
2
'2
,,
,
,,
,
,
lk
lk
l
k
Ec
x
y
Ec
x
y
E
H
x
y
W
x
y
,,
2,
,
,
lk
lk
E
cx
y
H
x
y
W
x
y
(20)
Acco
rdi
ng to the assu
mption
2
1
W
，
,
,,
0
lk
Ec
x
y
E
W
x
y
, and
,
Wx
y
is
irrelated with
,
,
lk
cx
y
with each
other, so:
,,
2
2'
,
2
11
,,
1
,
lk
l
k
MN
Bl
k
xy
lk
lk
E
cx
y
MN
(21)
Adopting the
unbia
s
e
d
esti
mation of
2
B
:
2
2'
,
2
11
1
,
MN
Bl
k
xy
cx
y
MN
(22)
From
whi
c
h i
t
is rea
d
ily seen that case B is the worse case, si
nce the
high
er the
variance the
higher the erro
r probability. So, we get:
2
1
2
2
f
B
Pe
r
f
c
(23)
Whe
n
8
10
f
P
, we get:
2
3.
9
7
2
B
(24)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel
HVS-ba
sed
Wate
rm
arking Sche
m
e
in Contou
rlet Tra
n
sfo
r
m
Dom
a
in (Hongb
o BI)
7523
5. Experimental Re
sults
The alg
o
rith
m ha
s bee
n
extensively tested
on vari
ous
stan
dard
image
s an
d
different
kinds of attacks. Some of t
he most significant results will
be shown. For the experiments
pre
s
ente
d
in the followi
ng, the host ima
g
e
is “L
ena
” wi
th the size of 512*5
12, the
false ala
r
m i
s
10
-8
, CT ex
ploits “9-7”
pyramid filte
r
and
“p
kva
”
directio
nal
filter to ob
tain a 2-l
e
vel
decompo
sitio
n
. First, wate
rmark invi
sibili
ty is val
uated
. In Figu
re 5,
the ori
g
inal
“L
ena” imag
e i
s
pres
ented on the left, while the
wa
termarked
c
o
py is s
h
own on the right. We
can
s
ee that the
image
s a
r
e
e
v
idently undi
stinguishable,
proving
the e
ffectiveness of
CT
wate
rm
arki
ng and
th
e
masking p
r
o
c
edure.
Figure 5. Ho
st Image (left) and the
W
a
te
rmarke
d Ima
ge (ri
ght) un
d
e
r No
Atta
cks
Other
re
sults after attacks a
r
e li
sted i
n
T
abl
e 1. It is ap
pa
rent
that the pro
posed
scheme i
s
re
sista
n
t to many signal o
p
e
r
ation.
Table 1. Ro
b
u
stne
ss Experimental
Re
sults
Attacks
par
ameter
s
PSNR
Detection
Results
noise
Peper & Salt
0.001
35.54
Y
e
s
0.010
25.43
Y
e
s
0.020
22.57
Y
e
s
G
auss
0.001
30.00
Y
e
s
0.003
25.22
Y
e
s
JPEG
Compr
e
ssion
70%
37.29
Y
e
s
50%
35.78
Y
e
s
30%
34.26
Y
e
s
Filtering
Average
3*3 33.75
Y
e
s
6*6 27.91
Y
e
s
Median
3*3 35.41
Y
e
s
6*6 27.77
Y
e
s
Geomet
ric
Scaling
25%
28.58
Y
e
s
75%
38.64
Y
e
s
150%
44.77
Y
e
s
Cropping
15%
21.79
Y
e
s
25%
18.80
Y
e
s
6. Conclusio
n
In this pape
r, a novel imag
e watermarki
ng schem
e has bee
n pre
s
ented. The al
gorithm
embed
s the
watermark code
by mo
difying the
CT
coeffici
ents
o
f
the
ima
ge, and exploits a
model a
dapti
ng the
wate
rmark st
ren
g
th to the
ch
a
r
acteri
stics
of the HVS. Th
e method
u
s
es
freque
ncy, lu
minan
ce an
d
texture anal
ysis to m
ode
l the HVS chara
c
teri
stics, increa
sing th
e
watermark strength without
gre
a
t
pe
rcep
tible disto
r
tio
n
. The
experi
m
ental results
sho
w
that t
h
e
prop
osed me
thod is
rob
u
st again
s
t ma
ny sign
al
pro
c
e
ssi
ng atta
cks
and th
e b
ehavior
of the
watermark de
tector
wa
s go
od.
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 12, Dece
mb
er 201
3: 751
6 – 7524
7524
Ackn
o
w
l
e
dg
ement
This
work i
s
sup
porte
d by
the Scien
c
e
& Te
chn
o
log
y
Project of Heilo
ngjian
g
provin
ce
unde
r Grant No. 125
210
5
6
.
Referen
ces
[1]
Ingemar J
Co
x, Joe Kil
i
a
n
, F
T
homson Lei
g
h
to, et al. Sec
u
re Spr
e
a
d
Sp
ectrum W
a
ter
m
arking
for
Multimed
ia.
IEEE Transactions On Im
age Processing
. 19
9
7
; 6(12): 16
73-
168
7.
[2]
Scott Craver,
Nasir M
e
mo
n, Boo
n
-Lock
Y
eo, et
al. R
e
s
o
lvin
g r
i
ghtful
o
w
n
e
rshi
ps
w
i
t
h
i
n
visi
bl
e
w
a
t
e
rmarki
ng t
e
chn
i
qu
es: Lim
i
tations, attack
s, and impl
icati
ons.
IEEE Journal On Selected Areas In
Co
mmun
icati
o
ns
. 1998; 1
6
(4)
:
573–5
86.
[3]
Qiming Liu. An Adap
tive Blind Waterma
rking Algorithm f
o
r
Color
Image.
T
E
LKOMNIKA Indo
nesi
a
n
Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 201
3; 11(1)
: 302-30
9.
[4]
Dee
pa Ku
ndur
, Dimitrios Ha
tzinakos. Div
e
r
sit
y
a
nd attac
k
characteriz
a
tion for impr
ov
ed rob
u
st
w
a
t
e
rmarking.
IEEE Transactions On Signal
Processing.
20
01; 49(1
0
): 238
3-23
96.
[5]
Humin
g
Gao, Li
yu
an Jia, Me
ilin
g Li
u. A Dig
ital
Watermarki
ng Alg
o
rithm for Col
o
r Image
Based 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.
[6]
Hon
gbo BI, Yu
bo ZHANG, X
u
emin
g LI. Vide
o
w
a
t
e
rmarki
n
g robust ag
ai
n
s
t spatio-temp
o
ral attacks.
Journ
a
l of Net
w
orks
. 2011; 6(6): 932-9
36.
[7]
Braci S, Bo
yer
R, Del
pha
C.
Analys
is of the
resistanc
e of the
Spr
e
a
d
T
r
a
n
sform
ag
ainst
T
e
mp
ora
l
F
r
ame Avera
g
i
ng
attack
. Inter
natio
nal
Co
nfe
r
ence
on Ima
g
e
Process
i
n
g
. Hon
g
Kon
g
. 2
0
10; 1: 2
13-
216.
[8]
Yubo Z
H
ANG,
Hon
gbo BI. T
r
ansp
a
rent Vi
de
o W
a
termarkin
g Explo
i
tin
g
Sp
atio-T
emporal
Maskin
g
i
n
3D-DCT
Doma
in.
Journ
a
l of C
o
mputati
o
n
a
l
a
nd Infor
m
atio
n
Systems
. 20
11
; 7(5): 1706-1
7
13.
[9]
MN Do, Martin Vetterli.
T
he Contour
let T
r
ansform: An Ef
fi
cient Directi
o
nal Mu
ltireso
l
u
t
ion Imag
e
Repr
esentati
o
n
.
IEEE
Transac
tions On Im
age Processing
. 2005; 14(
12): 20
91-2
106.
[10]
Shiva Z
a
bo
li,
M Shahr
am M
o
in.
CEW
:
A N
on-Bli
nd A
d
a
p
tive Imag
e W
a
termarkin
g Ap
p
r
oach B
a
se
d
On Entropy
in
Conto
u
rl
et Do
ma
in
.
IEEE
Internati
ona
l S
y
mposi
u
m o
n
Industri
a
l
El
ectronics. Vi
g
o
.
200
7; 1: 1687-
169
2.
[11]
Moha
n BC, Kumar SS. Rob
u
st digit
a
l
w
a
t
e
rmarki
n
g scheme
us
ing co
ntour
let
transform.
Internatio
na
l Journ
a
l of Co
mputer Scie
nce
and N
e
tw
ork Security
. 200
8; 8(2): 43-5
1
.
[1
2
]
H
a
i
f
en
g
Li
, We
i
w
ei
Son
g
,
Shux
un
Wa
ng
. R
o
b
u
s
t ima
ge
w
a
t
e
rmarkin
g al
go
rithm base
d
o
n
contour
let
transform.
Jour
nal o
n
Co
mmu
n
icati
ons
. 20
06
; 27(4): 87-94.
[13]
Shan
gqi
n Xia
o
, Hefei Lin
g
, F
uhao Z
ou, et al.
Adaptive Imag
e W
a
termark
i
n
g
Algorit
hm i
n
Conto
u
rle
t
Do
mai
n
. T
he
Japa
n-Ch
in
a Joint W
o
rksho
p
on
F
r
onti
e
r
of
Comp
uter
S
c
ienc
e
an
d
Te
ch
no
l
ogy
.
W
u
Han. 20
07;
1: 125-1
30.
[14]
Barni M, Barto
lini F
,
Pive A.
Improved
w
a
ve
let-b
a
sed
watermarki
ng th
roug
h pi
xel-
w
i
se maskin
g
.
IEEE Transactions on I
m
age
Processing
. 20
01; 5(10): 7
83–
91.
[15]
Hazem
A. Al-
Otum, Allam O
.
Al-T
aba’a. A
dapt
iv
e c
o
lor
i
m
age
w
a
term
a
r
king
bas
ed
o
n
a m
odifi
ed
improve
d
pi
xel
-
w
i
se maski
ng
techniq
ue.
Co
mp
uters an
d Electrical E
ngi
n
eeri
n
g
. 20
09; 35(5): 67
3-
695.
Evaluation Warning : The document was created with Spire.PDF for Python.