TELKOM
NIKA
, Vol.11, No
.11, Novemb
er 201
3, pp. 6281
~6
289
e-ISSN: 2087
-278X
6281
Re
cei
v
ed Ma
rch 2
9
, 2013;
Re
vised Jul 2
,
2013; Accep
t
ed Jul 20, 20
13
A Novel Self-adaptive Discrete Wavelet Transform
Digital Watermarking Algorithm
Chi Ma
1,2
*, Yong
y
ong Zhu
3
1
Colle
ge of Sof
t
w
ar
e, Univ
ersi
t
y
of Scienc
e a
nd T
e
chnol
og
y Liao
Nin
g, Ans
han, Ch
in
a
2
Dong
lin
g Sch
ool of Econ
omi
cs and Man
a
g
e
ment, Univ
er
s
i
t
y
of Scie
nce a
nd T
e
chnol
og
y Beiji
ng, Beij
in
g,
Chin
a
3
Departme
n
t of Economics a
n
d
Busin
e
ss Ad
ministratio
n
,
C
hon
gqi
ng U
n
iv
ersit
y
of Ed
uca
t
ion, Cho
n
g
q
in
g,
Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: asmachi@
12
6.com
A
b
st
r
a
ct
On the
b
a
sis
of th
e res
ear
ch of w
a
v
e
let
trans
fo
rm an
d d
i
g
i
ta
l
w
a
ter
m
ark
i
ng
te
chnology, this
pap
er pro
pose
d
a self-a
dapti
v
e discrete w
a
velet trans
for
m
(DW
T
) digital
w
a
termarki
ng
algor
ith
m
, w
h
ic
h
can ac
hi
eve th
e pur
pos
e of
e
m
b
e
d
d
in
g h
i
d
d
en w
a
ter
m
arks
by d
e
co
mp
os
e thre
e-lev
e
l w
a
vel
e
t of i
m
a
g
e
and d
e
co
mpos
e bit-pl
an
e of w
a
termarki
ng
gray scal
e
i
m
a
ge by Arn
o
ld s
c
rambli
ng tran
sformati
on. L
a
ye
r
ada
ptive thr
e
s
hol
d a
nd
qu
ant
i
z
e
r
w
e
re r
e
fer
ence
d
i
n
th
is
a
l
gorith
m
,
and
w
h
ich
ad
aptiv
e s
e
lecte
d
co
effici
ent
of deta
il s
u
b
b
a
nds
of e
m
b
e
d
d
ed w
a
ter
m
ark
i
ng to
i
m
pr
ov
e t
he r
obustn
ess
of the w
a
ter
m
a
r
king. In
testin
g
of
semi-bl
i
n
d
w
a
termarkin
g, ren
e
w
i
ng of w
a
termark
i
ng
base
d
on the e
m
b
e
d
d
in
g seq
uenc
e
of point loc
a
ti
ons
and th
e qu
anti
z
e
r
s
equ
enc
e
w
i
thout parti
ci
p
a
tion of the
ori
g
in
al i
m
a
ge. E
x
peri
m
e
n
tal re
sults show
that the
a
l
go
ri
thm
i
s
e
ffe
cti
v
e
to imp
r
ove
th
e rob
u
s
tne
ss o
f
th
e
cu
t,
a
d
d
i
n
g no
i
s
e
,
fi
l
t
e
r
i
n
g
,
an
d com
p
re
ssi
on
im
age
attack treatm
e
nt.
Ke
y
w
ords
:
di
g
i
tal w
a
termarki
ng, DW
T
,
arnol
d scrambli
ng, b
i
t plan d
e
co
mp
ositio
n
Copy
right
©
2013 Un
ive
r
sita
s Ah
mad
Dah
l
an
. All rig
h
t
s r
ese
rved
.
1. Introduc
tion
With the digi
tization of inf
o
rmatio
n and
the
flourish
of Internet, d
i
gital pro
d
u
c
ts have
become g
r
e
a
t
ly enriched
and e
a
sy to
sp
rea
d
, co
p
y
right protect
i
on an
d info
rmation
se
curity
issue
s
be
com
e
more p
r
omi
nent. Du
e to t
he defe
c
ts of
traditional i
n
formatio
n security tech
nolo
g
y
in digital produ
cts copyright prote
c
ti
on exists
,
contribute
d
to
the develo
p
ment of di
gita
l
watermarkin
g
tech
nolo
g
y. Digital
wate
rmarking
te
ch
nology
hide
s
the digital
wa
termark in di
gital
media, in ord
e
r to provide
copyri
ght ce
rt
ificates
for
co
pyright owners in co
pyright
dispute
s
. As an
effective mea
n
s to re
solve
copyrig
h
t issues of
digital
produ
cts h
a
s
bee
n wide
sprea
d
co
ncern.
Becau
s
e
of the p
r
oximity of the wavel
e
t transfo
rm a
nd hu
man vi
sual
syste
m
’s cha
r
a
c
ters, t
he
watermarkin
g
techniq
ue b
a
se
d on wavelet tran
sfor
m to becom
e
a research
hotsp
ot [1, 2]. So
far, the dom
estic a
nd foreign sch
o
lars have pr
op
ose
d
quite a
lot of digital waterm
arki
ng
algorith
m
ba
sed on
wavele
t transfo
rm. T
h
rou
gh th
e re
peated
emb
e
dding
and
u
s
e of Referen
c
e
watermarkin
g
method, the
referen
c
e [3]
achi
eved a
hi
gher dete
c
tio
n
rate.
Ho
we
ver, due to
the
referen
c
e waterma
r
k
addin
g
greatly in
creasi
ng
the do
ping amo
unt, thereby re
du
cing the im
ag
e
quality. Thro
ugh qualitativ
e and qua
ntitative analys
i
s
, the refere
nce [4] pro
p
o
se
d the wa
vele
t
transfo
rm d
o
m
ain imag
e
watermark e
m
beddi
ng
strategy and al
gorithm. Th
e
referen
c
e [5]
reali
z
ed
wate
rmark
ada
ptive embe
ddin
g
and
dete
c
ti
on with
out th
e ori
g
inal i
m
age, the
co
st
is
embed th
e bi
t is not fixed, and the ne
e
d
to reco
rd the emb
eddin
g
positio
n. Accordi
ng to the
embed
ded
waterma
r
k
cap
a
city and the
local ch
arac
teristics of the image, the
refere
nce [6]
dynamically adjuste
d the quantization st
ep size,
and
effectively enhan
ced t
he robu
stne
ss of the
algorith
m
.
On the ba
si
s
of the re
sea
r
ch of wavelet
transfo
rm a
n
d
digital wate
rmarkin
g
tech
nolog
y
[7-13], thi
s
p
aper p
r
op
oses
aself-ad
a
p
tive digital
watermarkin
g
algo
rithm
b
a
se
d o
n
inte
ger
wavelet tran
sform
a
tion
d
o
main. Algo
rithm achi
eve
d
the p
r
etre
atment of t
he
waterm
ark
informatio
n b
y
using
multi-scale
wavelet
tran
sf
orm
te
chn
o
logy, Arnold
scram
b
li
ng a
nd
bit pl
ane
decompo
sitio
n
techn
o
logy,
and the wat
e
rma
r
k h
a
s
stronge
r con
c
ealment. Also
in the algorit
hm
referen
c
ed la
yer ada
ptive thre
shol
d an
d quanti
z
ati
on f
a
ctor, a
daptiv
e sele
cted th
e coeffi
cients
of
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
0
87-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 628
1 – 6289
6282
the
detail
s
sub
-
ba
nd which embe
d
ded
in wate
rm
ark a
dapti
v
ely. In recovery dete
ction,
watermark
re
covery
ca
n
be ba
se
d on
locatio
n
s
se
quen
ce
of the emb
eddin
g
point an
d t
he
quanti
z
ation factor se
que
nce
[14-
16] without
th
e partici
pation
of
the o
r
igin
al imag
e, a
n
d
achi
eved se
mi-blin
d wat
e
rma
r
k
dete
c
tion. Experi
m
ental re
sult
s sh
ow th
at the algorith
m
is
effective to i
m
prove th
e robu
stne
ss
of the cu
t, addi
ng noi
s
e, filte
r
ing, a
nd
co
mpre
ssion i
m
age
attack treatm
ent
2. Pretrea
t
m
e
nt of
the Di
gital Wa
term
ark
In order to raise the difficulty of being cra
c
ked a
n
d she
a
r re
si
sta
n
ce, pretreat
ment is
necessa
ry b
e
fore the
dig
i
tal watermarking e
m
be
dd
ing into the
host ima
g
e. In the pap
er,
bit
plane d
e
com
positio
n [8] and Arnold
scram
b
ling [
9
] are u
s
e
d
in wate
rma
r
kin
g
informa
t
ion
pretreatment.
Then the qu
antizatio
n
alg
o
rithm is
reali
z
ed b
a
sed on
that the gray scal
e
imag
e is
conve
r
ted int
o
binary imag
e by Bit plan
e deco
m
po
sit
i
on. The co
rrelation of wat
e
rma
r
ki
ng'
s p
i
xel
spa
c
e
can b
e
clea
red by
Arnold scra
m
bling, and
then the wat
e
rma
r
ki
ng im
age
s be
com
e
meanin
g
le
ss
and di
sorgani
zed, so that the digital waterma
r
ki
ng be
come m
o
re hi
dden.
2.1. Arnold Scrambling
Image scra
m
b
ling is an inf
o
rmatio
n encryption te
ch
n
o
logy, and it is also a p
r
etreatment
pro
c
e
s
s for i
n
formatio
n hi
ding. T
here
are
thre
e p
u
r
po
se
s
of scrambli
ng
wat
e
rma
r
k ima
g
e
:
Avoiding the
block
effect,
addin
g
a
key to e
n
sure
th
e security of
watermarkin
g
,
and
en
han
cing
the robu
stne
ss of wate
rmarking. Arn
o
ld tran
sf
ormation is a
scramblin
g
technolo
gy that
comm
only used in digital i
m
age, comm
only kno
w
n a
s
cat face tra
n
sformation, the layout of the
gray value
in
the imag
e can be
ch
ang
ed by the
ch
ange
of the
pixel co
ordi
n
a
tes. If the di
gital
image i
s
viewed a
s
a matri
x
, then the image
will be
co
me "a me
ss",
but if continu
e
to use A
r
no
ld
transfo
rmatio
n, there
will
b
e
a
sam
e
im
a
ge
with
the
original im
age.
A new Arn
o
ld
tran
sformatio
n
for scram
b
lin
g watermarki
ng image is
adopte
d
her
e
,
the formulas of transfo
rm and inverse
trans
form are.
(
1
)
(
2
)
K and N
co
n
s
titute the pa
irs
(K, N) i
n
formul
a (1
), a
nd the p
a
irs
can
be a
s
th
e key of
scram
b
ling,
we ca
n u
s
e the
comm
on e
n
cryption alg
o
ri
t
h
ms
(such as DES) to encrypt it. Only th
e
person
who
maste
r
s the
key can resto
r
e the extract
e
d wate
rma
r
king to the ori
g
inal informat
ion,
so it
enh
an
ce
d the
se
cu
rity of
watermarking. In
ad
dition, the
ne
w
cat tran
sform
a
tion h
a
s ma
ny
advantag
es: it is unnecessary to calcul
ate the c
onv
ersi
on cy
cle of resto
r
ing the image. If
the
origin
al ima
g
e
is converte
d to a
state
a
fter N st
e
p
s,
then the
alg
o
r
ithm
can
re
store the
o
r
igi
n
al
image from t
he scra
mblin
g state by the same
step
s. Image can
be re
stored to any times, and
the efficien
cy
of recovery i
s
gr
eatly improved compa
r
ed with th
e cat transfo
rm,
it also
can
sa
ve
a lot of time.
Figure1 is the
Lena imag
e scram
b
ling.
(a) L
ena Ima
g
e
(b) n
=
1
(c
) n=
2
Figure 1. Re
sults of Arnold
Scram
b
ling
Digital
Image (n rep
r
e
s
ent
s the numb
e
r of
iterations)
11
mo
d
1
n
xx
N
yk
k
y
11
mo
d
11
n
xx
N
yk
y
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel Self-adapti
v
e Di
screte Wavelet Tran
sfo
r
m
Digital Waterm
arki
ng Algo
rithm
(Chi Ma)
6283
2.2. Bit Plane Decomp
os
ition
Image bit-pla
ne is
com
m
o
n
in imag
e codi
ng
and im
age
comp
re
ssion. Th
ere a
r
e ma
ny
kind
s
of represe
n
tation
s f
o
r
digital im
a
ge, take
g
r
ay
scale i
m
age
s as an
exam
ple, ea
ch
pix
e
l of
digital image
is co
nstituted
by the way of multi-bi
t, each pixel is u
s
ually 8 bits (i.e., 8 bits planes,
each bit is 0 or 1). Th
e meanin
g
of the bit planes
i
s
that decomp
o
sin
g
the gra
y
scal
e
value
of
pixel into
bin
a
ry value, th
en all
the
bits
(0 o
r
1)
wi
th the
same
value formed
the pl
ane. F
o
r
example, in a
n
image
whi
c
h gray
scale v
a
lue i
s
256,
e
a
ch
pixel re
prese
n
ted by a
singl
e byte, the
8 bina
ry bit
s
are B
7
B6B5B
4
B3B2B1B0
arrang
ed fro
m
high
to lo
w. In that way,
the B0
bit of all
the pixels co
nstituting the
0 bit-plan
e, B0 bit c
onstitut
i
ng the 1 bit-plane an
d so
on. As is sh
o
w
n
in Figure1, the image shou
ld contai
n 8 b
i
t planes..
Figure 2. Sch
e
matic Di
ag
ram of Bit Pla
n
e De
com
p
o
s
ition
grayscal
e im
age
plane 7
plane 6
plane 5
plane 4
plane 3
plane 2
plane 1
plane 0
Figure 3. Wat
e
rma
r
k Ima
g
e
Each Bit-pl
ane after Decompo
s
ition
The emb
edd
ed wate
rma
r
k imag
e is
grayscale imag
e in this pa
p
e
r. In ord
e
r t
o
achi
eve the g
oal of qu
antifying the wavelet co
effi
cie
n
ts of ho
st im
age throug
h
watermark va
lue
of 0
and
1 i
n
the
embe
dd
ed al
go
rithm, the
gray
scal
e waterm
ark
image
bit-pl
a
ne i
s
partitio
n
ed
into eight bin
a
ry image
s, a
s
sh
own in Figure 3.
3.Wa
termar
k
Embedding and Extr
acti
on Algorith
m Descrip
tio
n
3.1. Watermark Embeddi
ng Algorith
m
Let the h
o
st i
m
age
be
a
MN
gray scal
e ima
ge, an
d wate
rmark i
m
ag
e
be a
mn
gray scale im
age. This me
thod achi
eve
s
the pur
po
se of embeddi
ng a waterm
ark by qua
ntifying
the high-f
r
eq
uen
cy coeffici
ents, and t
he
spe
c
ific
step
s are as follo
ws.
1) A
c
hieve th
ree
-
level
wav
e
let de
com
p
o
s
ition of th
e o
r
iginal
ho
st i
m
age. L
e
t b
e
the L
-
th
layer de
co
mp
osition
of the
high-f
r
eq
uen
cy co
mpo
n
en
t,
,,
kh
v
d
den
ote h
o
r
izo
n
tal, verti
c
al,
diago
nal dire
ction compo
n
ent respe
c
tively. Condu
ct
one-dime
n
sio
n
al scan of th
e details of sub-
band
coeffici
ents from the
three directio
ns, re
spe
c
tive
ly, from high
scal
e
level to low scale lev
e
l,
and ge
nerate
three one
-di
m
ensi
onal
se
quen
ce
s of HL, LH, and HH.
32
32
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 628
1 – 6289
6284
2) Fi
rst
co
nd
uct Arnold
scrambli
ng, ge
n
e
rate
a two-d
i
mensi
onal
waterma
r
k ima
ge after
hashing g
r
ay
watermark im
age in o
r
de
r t
o
improv
e th
e wate
rma
r
k i
n
visibility. Then condu
ct bi
t-
plane
de
com
positio
n, from
high l
e
vel to
low l
e
ve
l, scan bit-plan
es at on
ce, thu
s
forming
on
e-
dimen
s
ion
a
l seque
nce of 0 and 1.
3) Acco
rdin
g
to the cha
r
acte
ri
stics o
f
wavelet co
efficients, small low-scal
e level
quanti
z
ation f
a
ctor a
nd le
ss se
gmentati
on; high-sc
al
e level quanti
z
ation
with bi
g factor a
nd
more
segm
entation
to generate quant
ified
factor sequ
ence.
1
2
Q
,,
3
4
Q
a
r
e th
e
spe
c
ific valu
e
s
of the thre
e layers. Wh
en dete
c
ti
ng,
due to the
small step
-size, the cha
n
g
e
in
quanti
z
ation
i
s
n
o
t si
gnifica
nt and
tend
s t
o
cau
s
e
dev
i
a
tions,
so
it is of ne
ed to
set the th
re
sh
old
value of the step-si
ze
s. Sp
ecific
step
-si
z
es a
r
e: level
1:
13
.
5
st
ep
; level 2:
23
.
5
step
; level 3:
35
ste
p
. Gene
rate a
threshold
val
ue sequ
en
ce
THR wi
t
h
t
h
e
sam
e
st
ep-
si
ze a
s
H,
V
,
a
nd D,
corre
s
p
ondin
g
to the
different level
s
of t
he
step-si
ze
t
h
re
shol
d. To
reflect
the
pri
o
rity pri
n
ci
ple
of
importa
nt co
efficient, it is sugg
ested t
o
try to
select a quantitative and
mean
ingful coeffici
ent.
Thus, the le
vel embedd
e
d
threshold
value is ada
ptively set.
De
cide the thre
shol
d value of
different level
s
a
c
co
rding
t
o
the l
e
vel in
whi
c
h
em
be
dded
coeffici
ent is,
and
th
e value
s
are
as
follows
res
p
ec
tively:
First level:
,
x
o
C
: the maximum in direction
o
,
1m
i
n
(
_
1
,
_
1
,
_
1
)
t
h
r
t
hr
h
t
hr
v
t
hr
d
;
Second level:
2
lo
g
2
_2
2
xo
C
thr
o
,
x
o
C
: the maximum in direction
o
,
2m
i
n
(
_
2
,
_
2
,
_
2
)
t
h
r
t
hr
h
t
hr
v
t
hr
d
;
Third level:
2
lo
g
2
_3
2
xo
C
th
r
o
,
x
o
C
: the maximum in direction
o
,
3m
i
n
(
_
3
,
_
3
,
_
3
)
t
h
r
t
hr
h
t
hr
v
t
hr
d
;
After gen
erat
ing all
the t
h
re
shol
d val
ues of
all
le
vel, a coefficient th
re
sho
l
d value
seq
uen
ce T
H
R is form
ed with the length of H, V, and D.
4) All the thresh
old value
s
are
set, in every level, given any
(,
)
mn
, s
o
rt
,
(,
)
hl
f
mn
,
,
(,
)
vl
f
mn
,
,
(,
)
dl
f
mn
from sm
all to big,
1,
(,
)
kl
f
mn
<
2,
(,
)
kl
f
mn
<
3,
(,
)
kl
f
mn
. Comp
ute
the
value of
step
-size
3,
1
,
(,
)
(
,
)
21
kl
k
l
f
mn
f
m
n
Q
, and
co
mpare
with
step, if
is
big
ger th
an
step, then go
on, or else skip to next point.
5) The spe
c
ific method fo
r quanti
z
ing th
e middle poi
n
t
is depicte
d
as Figu
re 4
Partition
the distan
ce of
3,
1
,
(,
)
(
,
)
kl
k
l
f
mn
f
m
n
, the num
ber o
f
intervals i
s
21
Q
, and
the si
ze of in
terval is
, the coo
r
din
a
te o
f
interval poi
nt is
()
Lj
,
([
0
,
2
1
]
)
jQ
. Com
p
ute
the location o
f
2,
(,
)
kl
f
mn
, and determine the qua
ntization valu
e of
2,
(,
)
kl
f
mn
accordin
g to the
watermark va
lue.
2,
(
,
)
[
(
2
)
,
(
2
1
)],
(
[
0
,
]
kl
f
m
n
Li
Li
i
Q
2,
(2
)
,
0
(,
)
(2
1
)
,
1
kl
Li
w
m
fm
n
Li
w
m
In orde
r to e
x
tract watermark, sto
r
e t
he
qu
antization facto
r
Q
of embed
poi
nt and its
locatio
n
into seque
nce EQ and IND re
sp
ectively.
Figure 4. Sch
e
matic Di
ag
ram of Quanti
z
ation Process
2
3
Q
2
log
2
_1
2
xo
C
th
r
o
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel Self-adapti
v
e Di
screte Wavelet Tran
sfo
r
m
Digital Waterm
arki
ng Algo
rithm
(Chi Ma)
6285
3.2. Watermark Extr
actio
n Algorithm
The algo
rith
m strictly limits the step
-si
z
e th
re
sh
old
value, coefficient threshol
d value,
and qua
ntizat
ion
fa
ctor
in watermark e
m
bed algo
rith
m, with th
e ai
m to imp
r
ove
the a
c
cu
ra
cy of
the extraction
process. Du
e to
limitations of the embedde
d wate
rm
ark, the extractio
n
accu
racy
of this alg
o
rit
h
m improves
a lot co
mpa
r
ed with th
e K
undu
r alg
o
rit
h
m. In additio
n
, this extra
c
t
i
on
algorith
m
ne
e
d
only be
em
bedd
ed the l
o
catio
n
s
equ
ence IND an
d qua
n
tize
d seque
nce EQ
[7],
without u
s
ing
the origin
al image
s.
1) Achieve th
ree
-
level
wav
e
let de
com
p
o
s
ition of th
e
detecte
d ima
ge, and
get t
he hig
h
-
freque
ncy co
mpone
nts in
hori
z
ontal, ve
rtical, diag
on
al dire
ction a
s
embe
dde
d algorith
m
.
2)
Con
d
u
c
t
an o
ppo
site
algo
rithm
o
f
embe
dded
wate
rma
r
k
to extra
c
t e
m
bedd
ed
watermark. Fi
nd the embe
d
ded lo
cation
according
to the embe
dde
d
location
seq
u
en
ce IND, sor
t
,
(,
)
hl
fm
n
,
,
(,
)
vl
fm
n
,
,
(,
)
dl
fm
n
of this point
,then get
1,
(,
)
kl
fm
n
<
2,
(,
)
kl
fm
n
<
3,
(,
)
kl
fm
n
.
3)
Ju
st a
s
the
method
whe
n
emb
eddi
ng,
partition th
e d
i
stan
ce
of
3,
1
,
(,
)
(
,
)
kl
k
l
fm
nf
m
n
,
with the interval numbe
r of
21
Q
, and the interval size of
3,
1
,
(,
)
(
,
)
21
kl
k
l
fm
n
f
m
n
Q
.
4) Find th
e a
pproxim
ation
interval point
of
2,
(,
)
kl
fm
n
,
2,
1
,
(,
)
(
,
)
()
kl
k
l
f
mn
f
m
n
ED
ro
und
.
5) If ED is
even, extrac
t the watermark
of this point a
s
0, or el
se a
s
1.
6)
Divide th
e on
e-dim
e
n
s
ion
a
l watermark
seq
uen
ce
of
wm
into eight bit-plan
e
s
according to
mn
.Then tra
n
sfo
r
m this into gray value, recover as
mn
grayscale waterm
ark
image.
4. Experimental Re
sults
and An
aly
s
is
The ba
sic
co
nfiguratio
n of the compute
r
in
experim
e
n
t is CPU co
re (TM
)
2 Du
o/2.5G,
memory of 4
G
, hard di
sk
of 360G. The
operatin
g
sy
stem is Wi
nd
ows 7, and the algo
rithm is
impleme
n
ted
using
simul
a
tion software Matlab 7.
The ori
g
inal
host imag
e in experim
ent
is
gray
scale i
m
age, a
nd th
e waterm
ark
image i
s
g
r
a
y
scal
e
ima
g
e
.
Achieve
wavelet d
e
co
mpositio
n on
host im
age
and
waterma
r
k im
age
usi
ng db
2 wave
let basi
s. During
Arnold h
a
shin
g pro
c
e
ssi
ng,
take
as 32,
as 1.
Embeddi
ng a
nd extra
c
ting
wate
rma
r
k i
n
the
ca
se
o
f
norm
a
l situ
ation with
out
attack
(sh
o
wn in Fi
g
u
re 3
)
, the ex
perim
ental re
sult sho
w
s th
at the emb
e
d
d
ed
waterma
r
k im
age i
s
still
intac
t, and the watermark
extrac
te
d
fro
m
emb
edde
d
image
is
also ba
sically co
nsi
s
tent
with
the
origin
al wate
rmark im
age.
From figu
re
(e)
we
ca
n
see th
e diffe
ren
c
e
s
bet
w
een the im
a
g
es
before
and
a
fter the wate
rmark e
m
be
d
d
ing. Thi
s
fig
u
re m
agnifie
s
the diffe
re
nce
s
of the t
w
o
image
s 10 times, depi
ctin
g the edge, conto
u
r outli
ne after emb
e
ddin
g wate
rmark into ho
st
image
u
s
ing
the al
gorith
m
, whi
c
h
is in
li
ne
with h
u
ma
n visu
al
syst
em, an
d
coul
d a
c
hieve
go
od
con
c
e
a
ling ef
fect.
a) Ori
g
inal Im
age an
d
Wate
rma
r
k I
m
age
(b) After Emb
eddin
g
Wate
rma
r
k a
nd Extracting
Wate
rma
r
k
(c) The T
w
o
Grap
hs
Differen
c
e E
m
bedd
ed
Watermark
before and after
NC=0.9
994,
PSNR=45.3
3
39db
Figure 3. Embeddi
ng and
Extracting Waterma
r
k in th
e Ca
se of No
rmal Situation
without Attack
512
512
32
32
N
K
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 628
1 – 6289
6286
(a)
White noi
se processe
d
image
(b) Salt and p
eppe
r noi
se p
r
ocesse
d ima
ge
Extracting wa
termarka
nd e
x
tracting waterma
r
k
NC=0.9
136, PSNR=39.9
4
66db
NC=0.8
620,
PSNR=22.323
4db
(c) Gau
s
sian
filter 3×3 p
r
o
c
essed ima
g
e
(d) G
a
u
ssi
an
filter 5×5 p
r
o
c
essed ima
ge
and
extracting
wa
termarka
nd e
x
tracting waterma
r
k
NC=0.8
626, PSNR=42.5
3
58db
NC=0.8
056,
PSNR=40.911
5db
(e) Me
dian filter processe
d image
(f) Enhan
ce
contra
st pro
c
e
s
sed imag
e a
nd
extracting
wa
termarka
nd e
x
tracting waterma
r
k
NC=0.8
376, PSNR=37.1
3
87db
NC=0.9
587,
PSNR=5.2574
db
(g)
Cut 1/16 p
r
ocesse
d ima
g
e
(h)
Cut 1/4 proce
s
sed ima
ge and extra
c
ting
watermarkext
ractin
g wate
rmark
NC=0.9
977, PSNR=16.0
7
68db
NC=0.9
565,
PSNR=10.063
7db
(i) Redu
ce
d to 1/4 pro
c
e
s
sed image a
n
d
extracting waterma
r
k
NC=0.8
759, PSNR=30.2
6
76db
Figure 4. Re
sult of Waterm
arked Imag
e after Adding
All Kinds of Attacks
In orde
r to test the robu
st
ness of the algorit
hm, attack su
ch a
s
noi
se, filtering, croppi
ng
is ad
ded into
the wate
rma
r
k imag
e. Figu
re 4
sho
w
s th
e value
s
of PSNR a
nd
NC detecte
d after
addin
g
all
ki
n
d
s
of atta
cks,
and
the
exp
e
rime
ntal
results d
e
mon
s
trate the
stron
g
robu
stne
ss of
the algorith
m
. Figure 5
sho
w
s the
Waterm
ark extraction i
m
age un
de
r different JPEG
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel Self-adapti
v
e Di
screte Wavelet Tran
sfo
r
m
Digital Waterm
arki
ng Algo
rithm
(Chi Ma)
6287
comp
re
ssion
ratio. Figure 6 and Fig
u
re 7 s
how
the values o
f
PSNR and
NC after JPEG
comp
re
s
s
ion at
t
a
ck.
Since thi
s
pa
per first conv
erts g
r
ayscal
e wa
te
rma
r
k i
m
age to a bi
nary seque
nce, thus it
is
c
o
ns
iderable to further t
e
s
t
the
extraction effec
t
by
cal
c
ulatin
g th
e bit e
rro
r
rati
o of waterm
a
r
k
extraction.
Define the bit
error
ratio
(BER)
of ori
g
inal
waterm
ark
and the extract
ed
watermark
as follows, in whi
c
h
denote watermark
seq
uen
ce le
n
g
th.
(
3
)
Q=
100
Q=
90
Q=
80
Q=
70
Q=
60
Q=
50
Q=
40
Figure 5. Wat
e
rma
r
k Extra
c
tion Image u
nder
Differe
nt JPEG Com
p
ressio
n Ratio
(Q re
prese
n
ts the comp
re
ssion ratio)
Table 1 de
pi
cts compa
r
i
s
on of the extracte
d
wate
rmark bit erro
r ratio of the prop
osed
algorith
m
an
d Kunde
r al
gorithm
after variou
s
atta
cks. It can
be seen tha
t
the extract
ed
watermark i
m
age
obtain
e
d
by the
propo
sed
alg
o
rithm
is
clo
s
e
r
to
the o
r
igin
al
waterma
r
k ima
ge
than the Kun
der alg
o
rithm,
and with a smaller bit
erro
r ratio an
d a stronge
r anti-a
ttack capa
bility.
Figure 8
sho
w
s th
e
comp
arison im
age
of extr
ac
ted watermark
bit error
ratio of the
prop
osed
alg
o
rithm
and K
unde
r al
gorit
hm after JP
EG co
mpression. It
sho
w
s that th
ere
i
s
a
signifi
cant im
provem
ent in
the rob
u
stn
e
ss of
the p
r
opo
sed al
go
rithm for JPE
G
com
p
ressi
o
n
attac
k
.
4. Conclusio
n
The p
r
o
p
o
s
e
d
alg
o
rithm
i
s
a
n
imp
r
ov
ement of
Ku
ndur qu
antization al
gorith
m
. The
Kundu
r algo
ri
thm first ada
ptively select
s digital imag
e discrete th
ree level wav
e
let coeffici
e
n
ts
,
and qua
ntifies co
rresp
ond
ing detailed
sub
-
ba
nd co
e
fficient acco
rding to the watermark va
lue.
Due to ela
b
o
r
ative sele
cti
on of quanti
z
ation in
terval
and embe
dd
ed location o
f
waterma
r
k, the
recovery effe
ct of digital
waterma
r
k is
g
ood, an
d ma
ke
s the d
e
tecting re
sult
s m
o
re int
u
itive. In
addition, it i
s
only of n
e
e
d
to em
bed
quanti
z
ation
seq
uen
ce i
n
to the
embe
d
ded
se
quen
ce to
recover
wat
e
rma
r
k
with
out parti
cipa
tion of origi
nal image, t
hus i
s
a ki
nd of semi
-blind
watermarkin
g
.
Experiment
al re
sults sh
o
w
that the
al
gorithm i
s
ro
bust to atta
ck operation
s
such
as noi
se, cut and comp
re
ssion.
Table 1. Co
m
pari
s
on of the
Extracted Waterma
r
k BE
R of the prop
ose
d
Algorith
m
and Kund
e
r
Algorithm after Various
Attack
s
A
t
t
a
c
k
s
prop
osed
alg
o
rithm
Kun
d
er
alg
o
rit
h
m
No Attack
0.0001
0.0000
White Noise
0.1023
0.1102
Salt and pepper
noise
0.0745
0.0873
Gaussian filter3×3
0.1769
0.1760
Gaussian filte
r
5×5
0.1772
0.1767
Median filte
r
0.2524
0.2881
Enhance contras
t
0.1395
0.1501
Cut 1/16
0.0040
0.0051
Cut 1/4
0.0293
0.0311
Reduced to 1/4
0.2497
0.3010
wm
'
wm
w
N
w
N
i
N
i
wm
i
wm
EBR
w
1
'
)
(
)
(
Evaluation Warning : The document was created with Spire.PDF for Python.
e-ISSN: 2
087-278X
TELKOM
NIKA
Vol. 11, No
. 11, Novemb
er 201
3: 628
1 – 6289
6288
Figure 6. The
Values of PSNR afte
r JPE
G
Compress
ion Attac
k
Figure 7. The
Values of NC after JPEG
Compress
ion Attac
k
Figure 8. Co
mpari
s
o
n
of Extracted Wa
termark
BER
after J
PEG Compress
ion
Ackn
o
w
l
e
dg
ement
This pa
per i
s
sup
p
o
r
ted
by The
National
Natu
ral
Scie
nce F
o
undatio
nof
China
No.
6120
2315.
Referen
ces
[1]
XD Z
h
ang, GD
Lu, J F
eng. T
he ima
ge co
di
ng bas
is an
d
w
a
vel
e
t compres
s
ion tech
no
log
y
- princ
i
p
l
es,
alg
o
rithms an
d
standards. Be
i
jing: T
s
inghu
a
Univers
i
t
y
Pr
es
s. 2004.
[2]
SG Li, GT
W
u
.
F
r
actal and W
a
velet, Beij
ing:
Scienc
e Press. 2002.
[3]
D
Ku
ndur, D Hatzin
akos.
At
tack char
acteri
z
a
ti
on
for
effective w
a
ter
m
ar
king
. Pr
oc of
Internatio
na
l
Confer
ence Im
age Proc
essin
g
. 1999; 2
40-2
44.
[4]
DR Hu
an
g, JF
Liu, JW
Hua
n
g
. An Embe
dd
ing
Strate
g
y
a
nd Al
gorithm f
o
r Image W
a
t
e
rmarkin
g i
n
DWT Domain.
Journ
a
l of Softw
are.
2002; 13
(7): 1290-
12
97
.
[5]
H Inoue, A Miyazaki, A T
a
kas
h
i.
Di
gital w
a
te
rmark
metho
d
usin
g the w
a
ve
let transfor
m
fo
r vide
o d
a
ta
.
Procee
din
g
s of
the 199
9 IEEE Internatio
n
a
l S
y
mp
osi
u
m on. 199
9; 4: 247-2
50.
[6]
YF
Shao, GW
W
u
, XG L
i
n.
Quantizat
i
on-
b
a
sed di
gital
w
a
termarki
ng al
gorithm.
Jou
r
na
l
o
f
Tsin
g
hua
Univers
i
ty (Science a
nd T
e
ch
nol
ogy)
. 20
03; 43(1): 20-
22.
[7]
CW
T
ang, HM Hang.
A
F
eature-Bas
e
d
Robust D
i
git
a
l Image W
a
termark
i
n
g
Sche
me
, IEEE
T
r
ansactions o
n
Processi
ng. 200
3; 51(4): 95
0-95
9.
[8]
CH F
e
i, D Kundur, RH K
w
ong. Ana
l
ysis
and des
i
gn of
secure
w
a
t
e
rmark-bas
ed authe
nticati
o
n
s
y
stems.
IEEE Trans on Infor
m
ation Forensics Security
. 20
06; 1(1): 43-5
5
.
[9]
Abdu
lja
bb
ar S
haam
ala, S
h
a
h
id
an M. A
bdu
llah
a
nd
Az
iza
h
A. Man
a
f, S
t
ud
y of th
e eff
e
ct DCT
an
d
DW
T
domains
on the im
perce
ptibil
it
y
and
r
o
b
u
stness of Gen
e
tic
w
a
termark
i
ng.
Internati
o
n
a
l Jour
nal
of
Co
mp
uter Scie
nce Issues
. 20
11; 8(5): 22
0-2
25.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
e-ISSN:
2087
-278X
A Novel Self-adapti
v
e Di
screte Wavelet Tran
sfo
r
m
Digital Waterm
arki
ng Algo
rithm
(Chi Ma)
6289
[10]
HHe, JP Z
han
g, Q W
ang. Digital Ima
ge W
a
termark Alg
o
r
i
thm Based
o
n
W
a
velet T
r
ansform an
d
Comp
oun
d Ch
aotic S
y
stem. J
ourn
a
l of h
a
r
b
in u
n
iv
ersity of scienc
e an
d
techno
logy
. 2
010; 1
5
(4): 15
-
18.
[11]
Y W
u
, Z
H
Gua
n
. A n
o
vel
di
git
a
l
w
a
t
e
rmark
al
gorithm
bas
ed
on c
haotic
map
s
.
Physics Letters A
. 20
07;
365(
11): 40
3-4
06.
[12]
Hu, DQ W
A
N
G, F
W
an. W
a
velet transform
w
a
termark
ing
method
bas
ed
on l
o
ca
l ph
as
e corre
latio
n
.
Appl
icatio
n Re
search of Co
mputers
,vol.2
8,n
o
.5,201
1,pp
19
22-1
925.
[13]
W
.
T
ang, H.M. Hang. A
Feature-B
a
se
d
Robust Dig
ital
Image W
a
termarkin
g Scheme,
IEEE
T
r
ansactio
n
s o
n
Processi
ng
. 200
3; 51(4): 95
0-95
9.
[14]
Rusdi
CY Cha
ng, HJ W
ang,
SW
Pan. A robust DWT
-
base
d
cop
y
ri
ght ver
i
ficatio
n
schem
e
w
i
t
h
F
u
zz
y
ART
.
Journal o
f
Systems and
Softw
are
. 2009
; 82(11): 19
06-
191
5.
[15]
DF
Ch
en, YQ
Z
han
g. Bl
ind
w
a
term
arkin
g
al
gor
ithm
b
a
s
ed
on
liftin
g
s
c
heme
w
a
v
e
l
e
t and
ch
aotic
mapp
ing.
Co
mputer Eng
i
n
eeri
ng an
d Des
i
gn
.
2007; 2
9
(20):
537
2-53
75.
[16]
Bao XHMA.
Image
a
d
a
p
tive w
a
t
e
rmarki
ng usin
g
w
a
ve
let doma
i
n sin
gul
a
r
value d
e
com
positi
on.
IEEE
T
r
ans on Circ
u
i
t
s and Systems for Video T
e
c
hno
logy
. 2
005;
15(1): 96-1
02.
Evaluation Warning : The document was created with Spire.PDF for Python.