Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
4, N
o
. 4
,
A
ugu
st
2014
, pp
. 57
3
~
58
4
I
S
SN
: 208
8-8
7
0
8
5
73
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
An Adaptive Steganography Sche
me Bas
e
d on Vi
sual Qu
ality
and Embedding Capacity Improvement
Mo
jt
ab
a B
a
h
m
anz
a
de
g
a
n
J
a
hr
omi
*
, K
a
ri
m Faez
*
*
* Departm
e
nt
of
Com
puter Engin
eering
,
S
c
ienc
e
a
nd Res
ear
ch Br
a
n
ch, Is
l
a
m
i
c A
z
a
d
Univers
i
t
y
,
Qa
zvin,
Iran
** Electr
i
cal
En
gineer
ing Dep
a
rtment, Amirkabi
r
University
of Technolog
y
,
Tehr
an,
I
r
an
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Apr 21, 2014
Rev
i
sed
May 30
, 20
14
Accepted
Jun 16, 2014
In this paper
,
a
steganograph
y
t
echniqu
e using
LSB substitutio
n and PVD
method is presen
ted
as an
adaptiv
e
sc
heme in
th
e spatial
domain. Our
method
partitions the gray
scale im
age in
to several non-o
v
erlapp
ing block
s
with three
cons
ecut
i
ve p
i
xe
ls
. Th
e em
bedd
ing algor
ithm can
both r
e
place the secret d
a
ta
with the LSBs
of the middle pixel a
nd embed
it in the differ
e
nce values
between
the middle pix
e
l and
its
two neighbor
in
g pixels of
th
e cover-block
.
The number of secret bits is dete
rmined ad
aptively
bas
e
d on the range
divisions for embedding in th
e diff
eren
ce v
a
lue. We d
e
fin
e
a new r
a
ng
e
division on gray level which
tak
e
s in
to accoun
t a
larg
er em
bedding
cap
a
c
i
t
y
for bits. After
th
e embedding
,
th
e propos
ed meth
od detects
the pixels which
are sensitiv
e to h
y
per distort
i
o
n
. Then, th
e e
m
bedding proce
ss will be
repeated to pro
duce insignificant visu
al distor
tion in those p
i
xels. Ou
r
experimental results dem
onstrate that
this iter
a
tive steg
anograp
h
y
schem
e
prevents significant visual distor
ti
on into stego-image. The g
e
nerated PSNR
values are high
er than
the
corres
pondi
ng valu
es
of the most com
m
only
used
methods, discussed in
this stud
y
.
Furthe
rmore, th
e experimen
t
al results show
that the hid
i
ng
capac
it
y
incr
ea
s
e
d e
normously
when the prop
osed range
division is used. Finally
,
we illust
rate th
at the method can pass R
S
and
s
t
egana
l
y
s
is
det
e
ctor
att
acks
.
Keyword:
Ada
p
tive stega
n
ography
Least-sign
ifican
t-b
it
su
bstitu
tio
n
Pi
xel
-
val
u
e di
f
f
ere
n
ci
n
g
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Mojta
b
a Bahm
anzade
g
a
n
Ja
hrom
i,
Facul
t
y
o
f
C
o
m
put
er an
d I
n
f
o
rm
at
i
on Tec
h
nol
ogy
En
gi
ne
eri
n
g,
Q
a
z
v
in
Br
an
ch I
s
la
mi
c Azad
Uni
v
ersity,
Baraj
i
n Un
iversity Ro
ad
,
Qazv
in
, Iran.
Em
a
il: b
a
h
m
an
zad
eg
an@q
iau.ac.ir
1.
INTRODUCTION
In
recent years
,
data sec
u
rity has becom
e
one of t
h
e m
o
st im
portant issue
s
of
hum
a
n societies due t
o
i
n
crease
d
dat
a
t
r
ansm
i
ssi
on o
v
er c
o
m
put
er
net
w
or
ks
[1]
.
I
n
t
h
i
s
c
o
nt
ext
,
secret
c
o
m
m
u
n
i
cat
i
on s
c
i
e
nc
e has
been prese
n
ted to inc
r
ease i
n
form
ati
on secu
r
i
t
y
. St
egan
og
ra
phy
i
s
o
n
e
of
t
h
e m
o
st
im
port
a
nt
t
ech
ni
q
u
es
use
d
t
o
pr
ovi
de safe
com
m
uni
cat
i
o
n an
d hi
de sec
r
et
m
e
ssages
[2]. It has bee
n
used since a
n
cient tim
e
s
and then
tu
rn
ed
i
n
to
an
i
n
tegral p
a
rt of
th
e d
i
g
ital era
after th
e
de
vel
opm
ent
of
com
put
e
r
s [
3
]
.
In
f
act
, st
egan
o
g
ra
phy
i
s
carri
ed
o
u
t
by
hi
di
n
g
sec
r
et
m
e
ssages i
n
t
o
a cove
r m
e
di
a such as t
e
xt
, i
m
age, vi
de
o, e
t
c. In t
h
i
s
pa
p
e
r, an
im
age has
bee
n
use
d
as
t
h
e
cove
r-i
m
a
ge a
n
d
t
h
e
res
u
l
t
i
m
age o
f
t
h
e e
m
beddi
ng
pr
o
cess i
s
nam
e
d st
eg
o-
im
age. In c
ont
rast, stega
n
alysis is the
scie
nce
o
f
fi
ndi
ng
s
u
c
h
hi
d
d
en
m
e
ssages
[4]
.
Visu
al
q
u
ality, em
b
e
d
d
i
ng
cap
acity, and
i
n
fo
rm
ati
o
n
secu
rity are t
h
ree features i
n
vestig
ated
b
y
researc
h
ers in steganogra
phic eval
uat
i
o
n.
M
o
re
ove
r, t
h
e pu
r
pose
o
f
st
egan
o
g
ra
phy
i
s
t
o
ren
d
e
r
secret
messag
e
s i
m
p
e
rcep
tib
le.
Alth
oug
h
it is i
m
p
o
s
sib
l
e to
ach
iev
e
ex
cellent resu
lts fo
r al
l th
ese featu
r
es in
a
steganogra
phy
schem
e
, acceptable levels can be realized
. Howe
ve
r, none
of them
shoul
d be prioritized ove
r
t
h
e ot
her
t
w
o.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
57
3
–
58
4
57
4
In a st
e
g
an
o
g
r
a
phy
m
e
t
hod,
t
h
e secret
m
e
ssage hi
di
n
g
pr
ocess i
n
v
o
l
v
es
t
w
o
basi
c st
eps. T
h
e fi
rs
t
st
ep i
s
em
bedd
i
ng a
nd t
h
e sec
o
n
d
i
s
e
x
t
r
act
i
n
g
.
I
n
t
h
e em
beddi
ng
p
h
ase,
a cove
r i
m
age i
s
chose
n
base
d o
n
t
h
e
st
egan
o
g
ra
phy
al
go
ri
t
h
m
perf
orm
a
nce, s
u
i
t
a
bl
y
fo
r
s
ecret data a
n
d the c
o
ver characte
r
istics. The
steganogra
phy schem
e
specifi
es ap
pr
op
ri
at
e regi
ons
o
f
t
h
e
cove
r-i
m
a
ge
and t
h
en em
beds the secret m
e
ssage
into them
. The
n
, t
h
e
resulting stego-im
age is forwarde
d to the
receive
r.
In the
e
x
tracting
pha
se, t
h
e
re
ceiver
g
i
v
e
s th
e steg
o to
th
e ex
t
r
actio
n fun
c
tio
n and
th
e secret m
e
ssag
e
will b
e
ex
tracted.
In vie
w
of
putt
i
ng secret data
into the cove
r-im
age, existing steganogra
phy
m
e
thods ca
n be divide
d
i
n
t
o
t
w
o
gene
r
a
l
gro
u
p
s:
(i
) Tran
sf
orm
do
m
a
i
n
m
e
t
hods,
(i
i
)
Spat
i
a
l
do
m
a
i
n
m
e
t
hods.
The t
r
ans
f
o
r
m
dom
ai
n
ap
pro
ach
es [5
]-
[9
] c
onv
ert the cov
e
r-im
ag
e in
to
ano
t
h
e
r
do
m
a
in
(lik
e the frequ
en
cy
) fi
rst, an
d th
en
em
b
e
d
t
h
e secret
m
e
ssage i
n
t
o
t
h
e
t
r
ansf
o
r
m
e
d coef
fi
ci
ent
s
.
A
l
t
hou
g
h
t
h
ese
m
e
t
hods a
d
e
q
uat
e
l
y
resi
st
agai
nst
st
eganal
y
s
i
s
at
t
acks, t
h
ei
r t
i
m
e com
p
l
e
xi
t
y
is hi
g
h
. I
n
p
r
o
p
o
se
d spat
i
a
l
do
m
a
i
n
m
e
t
hods [1
0]
-[
2
1
]
,
t
h
e secret
d
a
ta is em
b
e
dd
ed d
i
rectly in
to
th
e cov
e
r-p
i
x
e
l v
a
l
u
e.
Du
e to
th
ei
r low ti
me co
m
p
lex
ity
, th
ese m
e
th
o
d
s are
q
u
ite co
mm
o
n
.
Althou
gh
t
h
ese techn
i
qu
es p
r
ov
id
e
h
i
gh em
b
e
d
d
i
ng
cap
acity, th
e st
eg
o-im
ag
e q
u
a
lity i
s
co
nsu
m
ed
ly red
u
c
ed
at h
i
gher h
i
d
i
n
g
cap
a
cities. In
crease in
th
eir e
m
b
e
d
d
i
n
g
cap
acity wh
ile m
a
in
tai
n
ing
acceptable
qua
lity appears to
have
bec
o
m
e
a
challenge.
He
re,
we p
r
o
p
o
se
d a m
e
t
hod t
o
pr
o
v
i
d
e
go
o
d
s
t
eg
o
q
u
a
lity at h
i
g
h
er cap
acities.
Spatial dom
ain approac
h
es ca
n be divi
de
d into th
ree m
a
in categories: (i) LSB replacem
ent
m
e
thods
,
(ii) Edge ada
p
t
i
ve m
e
thods
, (i
ii) Hybri
d
m
e
thods.
In
LSB replacem
ent
m
e
thods, L
S
Bs
of pixels are
use
d
to
h
i
d
e
secret m
e
ssag
e
s
[1
0
]-[13
]. Th
e LSB su
b
s
titu
tion
is
a w
e
ll-kno
wn
tech
n
i
qu
e i
n
t
h
is gro
u
p
.
A
lthough
these m
e
thods
provi
de a large em
bedding capacity, they ar
e v
e
ry suscep
tib
le to
st
eganalytic atta
cks.
In
g
e
n
e
ral, eyes
d
e
tect ch
ang
e
s in
sm
o
o
t
h
areas with
add
ition
a
l cap
ab
ilitie
s th
an
ed
g
e
areas b
a
sed
on
h
u
m
an
vi
sual
sy
st
em
(H
VS
) cha
r
act
eri
s
t
i
c
s. The s
econ
d
t
y
pe
of
spat
i
a
l
dom
ain m
e
t
hods [
1
4]
-[
1
7
]
i
s
t
h
e edg
e
adapt
i
v
e
m
e
t
h
ods
usi
n
g
t
h
i
s
feat
u
r
e. T
h
e
PV
D m
e
t
hod
[1
4]
i
s
an e
x
a
m
pl
e of t
h
i
s
gr
o
u
p
.
Thi
s
m
e
t
h
o
d
com
put
es t
h
e secret
dat
a
em
beddi
ng ca
paci
t
y
by
t
h
e di
ffer
e
nce val
u
e
s
bet
w
een eac
h pi
x
e
l
and i
t
s
nei
g
h
b
o
r
i
n
g
p
i
x
e
ls.
Alth
oug
h
th
e ed
g
e
ad
ap
tiv
e m
e
th
od
s produ
ce h
i
g
h
v
i
su
al qu
ality,
th
eir h
i
d
i
n
g
cap
acity is
lo
w
com
p
ared with other techniques lik
e LSB and the like
.
Recently in [18]
-[21], the LSB
substitution and t
h
e
edge a
d
apt
i
v
e
m
e
t
hods
have
been c
o
m
b
i
n
ed. Th
us
, t
h
e t
h
i
r
d t
y
pe of s
p
a
t
i
a
l
do
m
a
i
n
m
e
t
h
o
d
s i
s
t
h
e h
y
b
ri
d
m
e
t
hods
whi
c
h hi
de i
n
bot
h LSB
s
an
d
di
ffe
re
nce val
u
es bet
w
een
nei
g
hb
o
r
i
n
g p
i
xel
s
. These a
r
e t
h
e
approaches
wi
th which
rese
arche
r
s
have
tried to obtain acceptable
va
lu
es for each of
t
h
e
three
feature
s
im
port
a
nt
i
n
t
h
e st
e
g
an
o
g
ra
p
h
i
c
eval
uat
i
o
n.
The
schem
e
p
r
o
p
o
sed
by
K
h
odaei
et
al.
[19] offe
rs acce
ptable
visual
quality and large em
beddi
ng capacity
of st
ego-im
age am
ong
othe
r
m
e
thods
in t
h
is category.
In t
h
i
s
pa
per
,
a hy
bri
d
t
ech
ni
q
u
e i
s
pro
p
o
se
d usi
n
g LSB
repl
acem
e
nt
and P
VD m
e
t
hod. T
h
e schem
e
b
o
t
h
enh
a
n
ces th
e h
i
d
i
n
g
cap
acity an
d
i
m
p
r
ov
es th
e
v
i
su
al q
u
a
lity at
h
i
gh
er cap
acities. Th
ree
d
i
fferen
t
adva
nt
age
s
of
ou
r sc
hem
e
i
n
com
p
ari
s
on
wi
t
h
ot
her
m
e
t
hods a
r
e as
f
o
l
l
o
ws:
1.
B
y
defi
ni
n
g
ne
w ra
nge
di
vi
si
on
on
gray
l
e
v
e
l
R
=
[0,2
5
5
]
,
we are abl
e
t
o
em
bed l
a
rge se
cret
bi
t
s
, hi
g
h
e
r
th
an
th
e Yan
g
et al.'
s
[
15]
,
t
h
e
Wu
et a
l
.'
s
[18] and t
h
e
Khodaei
et al
.'
s
[1
9]
tech
n
i
qu
es.
2.
Ou
r sc
hem
e
p
r
eve
n
t
s
l
a
r
g
e
di
ffe
re
nce
val
u
es
bet
w
ee
n the c
o
ver im
ag
e and its ste
g
o-im
age'
s pixe
ls,
unlike the Khodaei
et al.'
s
[
19]
t
ech
ni
q
u
e.
Thus
, i
t
pro
d
u
ces i
n
si
g
n
i
f
i
c
ant
vi
sual
di
st
ort
i
o
n i
n
hi
dd
en
m
e
ssages.
3.
Fin
a
lly, ou
r al
g
o
rith
m
resists ag
ainst RS steg
an
alysis attack
,
un
lik
e th
e
Wu
et al.'s
[
18] an
d
t
h
e
Y
a
ng
et
al
.
'
s
[
15]
m
e
t
hods
.
The rem
a
inder of this pa
per i
s
organize
d as follows
. In Sec
tion 2,
we re
view two
well-a
ccepted data
em
beddi
ng
sc
hem
e
s usi
ng t
h
e LSB
re
pl
ac
em
ent
and t
h
e
PVD m
e
t
hod
.
Then
, i
n
Sect
i
on
3, w
e
p
r
es
ent
o
u
r
p
r
op
o
s
ed
m
e
th
o
d
. Ex
p
e
rim
e
n
t
al resu
lts will b
e
d
e
scrib
e
d
an
d
co
m
p
ared
with
th
e
Yang
et al.'
s
[15
]
,
the
W
u
et
al
.
'
s
[
18]
a
n
d t
h
e
Kh
o
d
aei
et a
l
.'
s
[
19]
m
e
t
hods
i
n
Sect
i
o
n
4.
Fi
nal
l
y
, t
h
e
concl
u
si
o
n
fol
l
ows
i
n
Sect
i
o
n
5.
2.
ANALYSIS OF RELEVE
N
T APPROACHES
Here,
we
will d
e
scri
b
e
t
w
o
m
o
st co
mm
o
n
l
y u
s
ed
m
e
th
o
d
s in
t
h
e
n
e
x
t
su
b
s
ectio
n
s
. These techn
i
qu
es u
s
e th
e
com
b
ination
of LSB
replacem
ent and t
h
e P
V
D m
e
thod i
n
s
p
atial dom
ain.
2.
1.
Ste
gan
ogr
a
ph
ic Method usi
n
g LSB
Su
bsti
tuti
on
an
d P
V
D
Wu
et
al.
[
18]
pro
p
o
se
d a
st
egan
o
g
ra
phi
c
m
e
t
hod
fo
r
g
r
ay
scal
e im
ages i
n
2
0
0
5
.
S
u
p
p
o
se t
h
at
,
and
j
1
,2,
…
,5
whe
r
e
and
are t
h
e lower and
uppe
r
bound values a
n
d
|
|
is th
e len
g
t
h
of
t
h
e
ran
g
e.
Thi
s
a
p
pr
oach
di
vi
des
t
h
e R
=
[0
,2
5
5
]
ra
nge
t
o
t
h
e s
u
b-
ran
g
es
0,
15
,
16,
31
,
32
,
63
,
64
,
127
and
128
,
255
].
Th
e
div
defi
ne
s t
h
e
l
o
cat
i
on
of
ra
n
g
e di
vi
si
o
n
s
w
h
ere
ϵ15,31,63
,127
. Th
en, th
e sub-ran
g
e
s
wh
ich
are lower th
an
di
v
fall in
to
t
h
e ‘lower lev
e
l’ and
th
e o
t
h
e
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
An
ad
ap
tive st
eg
ano
graph
y sch
e
me ba
sed
on
visu
a
l
qu
a
lity an
d emb
e
d
d
i
n
g
capa
city …
(Mo
jta
ba
B.J.)
57
5
su
b-rang
es are lo
cated in
t
h
e ‘h
igh
e
r lev
e
l’. For i
n
stan
ce
, Fi
gu
re 1 sh
o
w
s
a di
vi
si
o
n
o
n
t
h
e gray
l
e
vel
w
h
en
31
.
At first, this m
e
thod pa
rtitions the cover-im
a
ge
into several non-ove
rlapping bl
ocks
having two
con
s
ecut
i
v
e
pi
xel
s
,
den
o
t
e
d b
y
(
,
). For
eac
h block,
the difference value
is calcu
lated
b
y
|
|
whe
r
e
i
i
s
t
h
e bl
oc
k
num
ber.
Aft
e
r
w
ar
ds
, t
w
o
scena
r
i
o
s m
a
y
occur i
n
t
h
e sec
r
et
d
a
t
a
em
bedd
i
n
g
pr
ocess:
Ca
se 1
(I
f the
diffe
re
nce val
u
e
falls in
to
th
e lo
wer lev
e
l): in
th
is case, th
e LSB rep
l
ace
m
e
n
t
m
e
t
hod i
s
use
d
t
o
em
bed
6bi
t
s
o
f
sec
r
et
dat
a
.
Ca
se 2
(
I
f
the
diffe
re
nce
valu
e
is placed
in the higher
leve
l):
the
PVD method is
use
d
t
o
em
be
d
the secret
bits.
In this case
,
t
h
e n
u
m
b
er o
f
e
m
bedded
bi
t
s
is calcu
lated
b
y
|
|
|
|
.
2.
2.
Ad
apti
ve S
t
e
g
an
ogr
ap
hi
c M
e
th
od u
s
i
n
g
L
S
B
Sub
sti
t
uti
o
n
an
d P
V
D
K
hod
aei
et al.
[1
9]
i
n
t
r
o
d
u
c
e
d an a
d
a
p
t
i
v
e
st
egan
o
g
ra
phi
c
m
e
t
hod
fo
r
gray
scal
e i
m
ages i
n
20
1
1
.
Su
pp
ose
t
h
at
,
and
j
1
,2
,
…
,
5
whe
r
e
and
ar
e t
h
e l
o
wer
an
d upp
er
b
oun
d v
a
l
u
es.
Th
e
l
e
ngt
h
of
i
s
de
not
e
d
by
|
|
.
As
dem
onst
r
at
ed
i
n
Fi
g
u
re
2
,
t
h
i
s
m
e
t
hod l
o
cat
es t
h
e
su
b
-ra
ng
es
0
,
7
,
7
,
1
5
,
16
,
3
1
,
32
,
63
and
64,
255
in
two
lev
e
ls, d
e
no
ted
b
y
‘lower lev
e
l’
and
‘hi
ghe
r l
e
vel
’
.
The
n
, i
t
defi
ne
s
Type
1
and
Typ
e
2
di
vi
si
o
n
s
on t
h
e gray
l
e
vel
.
I
n
t
h
e
Ty
pe1
di
vi
si
on s
h
ow
n i
n
Fi
gu
re 2a
, t
h
e
sub
-
r
a
n
g
es
0,
7
،
7,
15
and
16,
31
are locat
ed in the
‘lower level’ a
nd
the su
b
-ra
nge
s
32
,
63
and
64,
255
fall in
to
t
h
e ‘h
igh
e
r lev
e
l’. It
h
i
d
e
s
3
bits of secret
data
in the
s
u
b-ra
nges
of the
lower level
,
a
n
d c
onceals
4
bi
t
s
o
f
sec
r
et
dat
a
i
n
t
h
e
su
b-
ra
nge
s o
f
t
h
e
hi
g
h
e
r
lev
e
l. Also
, in
Type
2
di
vi
si
o
n
s
h
ow
n i
n
Fi
gu
re
2
b
, t
h
e s
u
b
-ra
n
g
es
0,
7
,
7
,
15
,
16
,
31
and
32,
63
b
e
lon
g
t
o
th
e
‘lower level’ an
d
64,
255
is assi
g
n
e
d to th
e
‘h
i
g
h
e
r lev
e
l’.
In th
is
di
vi
si
o
n
t
y
pe
, t
h
e num
ber of
secret
bi
t
s
is calcu
lated in
th
e lower lev
e
l, b
y
|
|
|
|
an
d in the
h
i
gh
er lev
e
l,
by
|
|
|
|
.
At first, this
m
e
thod
partitions
t
h
e cove
r-im
ages into se
veral non
-overlapping
bloc
ks
with three
con
s
ecut
i
v
e
pi
xel
s
, de
not
e
d
by
whe
r
e
i
is
the bl
ock number. T
h
e
k
and
ϵ3,4,
5
,6
defi
ne
t
h
e nu
m
b
er
o
f
secret b
its th
at can
b
e
em
b
e
dded
in
th
e LSBs.
Th
erefo
r
e, the e
m
b
e
d
d
i
n
g
cap
acity will g
r
o
w
as
k
is inc
r
eased.
The t
y
pe of ra
nge
di
vi
si
o
n
and t
h
e
k
val
u
e shoul
d
be selected in the da
t
a
em
beddi
n
g
p
r
oces
s fi
rst
.
Fo
r ea
c
h
cove
r
bl
ock
,
t
h
e
k
-b
its
o
f
secret d
a
ta are su
bstitu
ted
for t
h
e
k
-L
SB
o
f
m
i
ddl
e pi
xel
and
i
s
obt
ai
ned
.
Th
en
, th
e
v
a
lues
and
are calculated by
the differe
n
ce
betwee
n the
middle pixel
and i
t
s
t
w
o
nei
g
hb
o
r
i
n
g pi
xel
s
and
. The
and
rang
es
to
wh
ich
and
belong, a
r
e se
lected from
the
considere
d
ra
nge
division. T
h
e ne
w
differe
n
ce val
u
es
an
d
are
calculated by t
h
e
num
b
er
of secret bits
and
an
d t
h
e u
ppe
r an
d t
h
e l
o
wer
bo
u
n
d
s
o
f
t
h
ei
r ra
nge
s. Fi
nal
l
y
, t
a
ki
ng t
h
e new
di
f
f
ere
n
ce val
u
es
and
i
n
t
o
c
o
nsi
d
erat
i
o
n,
t
h
e
m
e
t
hod
p
r
o
d
u
c
e
s t
w
o
val
u
es
fo
r eac
h
pi
xel
.
Whereas
one
of
t
h
em
has m
u
c
h
diffe
re
nce c
o
mpare
d
with t
h
e
ori
g
inal
v
a
lu
e
o
f
cov
e
r-im
ag
e, th
e
similar v
a
lu
e is ch
o
s
en
fo
r its stego
-
imag
e.
Hig
her level=[32
,
25
5]
Lo
wer level=[0,31]
R
5
=[128,255]
R
4
=[64,127]
R
3
=[32,63]
R
2
=[16,31]
R
1
=[0,15]
Fi
gu
re
1.
A
n
e
x
am
pl
e of
ran
g
e
di
vi
si
on
o
n
g
r
ay
l
e
vel
(R
=
[
0
,
2
55]
)
i
n
t
o
‘l
o
w
er
l
e
vel
’
a
n
d
‘hi
ghe
r l
e
vel
’
i
n
t
h
e
Wu et
al.
m
e
t
hod
(
di
v
=31)
Higher lev
e
l=[3
2,255]
Lo
w
e
r level=[0
,31]
R
5
=[64,255]
R
4
=[32,63]
R
3
=[16,31]
R
2
=[8,15]
R
1
=[0,7]
(a)
Higher level=[64,255]
Lo
w
e
r level=[0
,63]
R
5
=[64,255]
R
4
=[32,63]
R
3
=[16,31]
R
2
=[8,15]
R
1
=[0,7]
(b
)
Fi
gu
re 2.
Tw
o ran
g
e di
vi
si
o
n
s
o
n
gray
l
e
vel
0
,
255
in
to
‘lower level’ and
‘h
i
g
h
e
r
lev
e
l’ in th
e
Khodaei et al
.
m
e
t
hod,
a
)
Ty
pe1
,
b
)
Ty
pe2
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
57
3
–
58
4
57
6
3.
THE PROPUSED
METHOD
In t
h
is section,
we will de
fine
our
propose
d
m
e
t
hod for gra
y
scale
im
ages.
The
aim
s
of proposi
ng t
h
is
m
e
thod a
r
e fi
ndi
ng se
nsitive pixels
to the hyper
distortions and e
x
te
ndi
ng t
h
e data
em
beddi
ng
proces
s
recursiv
ely.
Al
so
, t
h
e m
e
th
o
d
will b
e
increase th
e
h
i
d
i
ng
cap
acity th
ro
ugh
d
e
fi
n
i
ng
a
n
e
w rang
e
d
i
v
i
si
o
n
. Our
pr
o
pose
d
m
e
t
hod
i
s
prese
n
t
e
d i
n
t
h
ree
pha
ses:
(i
)
R
a
nge di
vi
si
o
n
o
n
gr
ay
l
e
vel
p
h
ase,
(ii
)
Dat
a
em
beddi
n
g
p
r
o
cess, (iii)
Data ex
tracting
p
r
o
cess, in th
e
fo
llowing
sub
s
ectio
n
s
.
3.
1.
R
a
ng
e D
i
v
i
sions on Gray
Lev
e
l Pha
s
e
Here
,
we
c
o
nsi
d
er
t
w
o ran
g
e di
vi
si
o
n
s on g
r
ay
l
e
vel
,
Type
1
and
Ty
pe
2
. The
Ty
pe
1
di
v
i
si
on i
s
u
s
e
d
to
co
m
p
are th
e p
r
o
p
o
s
ed
m
e
t
h
od
with
t
h
e oth
e
r related
m
e
th
od
s b
a
sed
o
n
th
e v
i
su
al qu
ality, sp
ecified
b
y
th
e
K
hod
aei
et al.
[19]
m
e
thod.
Additionally, the
Type
2
di
vi
s
i
on i
s
de
fi
ne
d as ou
r p
r
o
p
o
se
d ra
nge
di
vi
si
on i
n
or
der
t
o
achi
e
ve a
l
a
rge
r
em
bed
d
ing
capaci
ty. First, s
u
ppose t
h
at
,
and
j
1
,2,
…
,5
whe
r
e
and
are
t
h
e
l
o
wer
an
d u
ppe
r bo
u
n
d
val
u
es.
The wi
dt
h of
i
s
de
not
e
d
by
|
|
.
3.
1.
1.
T
y
pe 1 Di
vi
si
o
n
H
e
r
e
, t
h
e sub-r
a
ng
es
0,
7
,
7,
15
,
16,
31
and
32
,
6
3
are
put
in t
h
e
‘l
o
w
er l
e
vel
’
and t
h
e s
u
b
-ra
nge
64
,
255
in the
‘higher le
vel’,
according to t
h
e Khodaei
et
al.’s
di
vi
si
o
n
s
h
o
w
n i
n
Fi
gu
re 2
b
.
In t
h
i
s
t
y
pe
of
ran
g
e
di
vi
si
on
, t
h
e
num
ber o
f
secret
bi
t
s
is calcu
lated
b
y
|
|
||
in th
e sub-
r
a
ng
es
of
t
h
e l
o
wer
lev
e
l an
d
by
|
|
||
in th
e
h
i
gher lev
e
l.
So, the
num
ber of
t
h
e em
bedde
d bi
t
s
fo
r
wh
er
e
j
1
,2,
…
,5
will b
e
3
,
3
,
4
,
5
and
6
.
3.
1.
2.
T
y
pe 2 Di
vi
si
o
n
W
e
assi
gn
th
e
su
b-
rang
e
0,
7
to th
e
‘lower lev
e
l
’
, t
h
e sub
-
ranges
7,
15
,
16
,
31
and
32,
63
t
o
t
h
e
‘
m
i
ddl
e l
e
vel
’
,
an
d t
h
e s
u
b-
r
a
nge
64,
255
to
t
h
e
‘h
igh
e
r lev
e
l’,
in
th
e
pr
o
pose
d
Typ
e
2
di
vi
si
o
n
s
h
o
w
n i
n
Fi
gu
re
3.
Here
, t
h
e
num
ber
of
se
cret
bi
t
s
is calcu
lated
b
y
|
|
|
|
in th
e su
b-ran
g
e
o
f
th
e l
o
wer lev
e
l,
b
y
|
|
||
+1
i
n
th
e m
i
d
d
le lev
e
l, and
b
y
|
|
|
|
in
th
e
h
i
gh
er lev
e
l. Thu
s
, it will b
e
3
,
4
,
5
,
6
and
6
for
and
j
1
,2,
…
,
5
.
3.
2.
Data Embed
d
i
ng Pr
ocess
First, on
e o
f
the
Type1
or
t
h
e
Type
2
di
vi
si
on
s sho
u
l
d
be sel
ect
ed. The fl
o
w
di
ag
ram
of the pr
o
p
o
s
ed
st
egan
o
g
ra
phy
schem
e
i
s
il
lust
rat
e
d i
n
Fi
gu
re 4
.
W
e
s
u
g
g
est
1
2
st
e
p
s fo
r t
h
e
pr
o
p
o
se
d dat
a
em
beddi
ng
pr
ocess
i
n
t
h
e f
o
l
l
o
wi
ng
p
r
oce
d
u
r
e:
St
ep 1
:
A
g
r
ayscale i
m
ag
e is p
a
rtitio
n
e
d
in
to
non
-ov
e
rlap
p
i
n
g
b
o
c
k
s
hav
i
ng
th
ree con
s
ecu
tiv
e p
i
x
e
ls. Th
e
fi
rst
pi
xel
,
m
i
d
d
l
e
pi
xel
a
n
d s
econ
d
pi
xel
ar
e de
n
o
t
e
d
by
(
,
,
, w
h
ere
i
i
s
t
h
e
num
ber
o
f
t
h
e
bl
oc
k
.
S
also
denotes the secret
data.
St
ep 2:
C
o
n
s
id
e
r
th
e
k
value
where
ϵ3,4,
5
,6
is the num
b
er of the sec
r
et bits
that can
be e
m
bedded i
n
LSB
s
. T
h
e
n
, t
h
e em
beddi
n
g
c
a
paci
t
y
i
s
i
n
cre
a
sed
by
t
h
e
hi
g
h
er
val
u
e
of
k
. Thus,
is ob
tai
n
ed b
y
pu
ttin
g
k-
leftmo
s
t
bi
t
s
of
t
h
e
bi
na
ry
secr
et
dat
a
(
) i
n
to
k-
r
i
gh
tmo
s
t
b
its o
f
LSBs
of
.
St
ep 3
:
Com
p
ute the di
ffe
renc
e value
bet
w
e
e
n
and
using
Eq
. (1
).
(1
)
whe
r
e
is th
e
deci
m
a
l v
a
lu
e of
k
-
rightmost
LS
Bs
of
and
is
th
e d
ecim
a
l v
a
lu
e
o
f
k-leftmost
b
its of
.
St
ep 4
:
Use
opt
im
al
pi
xel
ad
ju
stm
e
nt
pr
ocess
(O
PA
P)
[
15]
a
n
d
al
t
e
r t
h
e
val
u
e
of
, as s
h
ow
n in E
q
.
(2).
H
i
gher level=
[
64,
255]
Lo
w
er lev
e
l=[0
,7
]
M
i
ddle lev
e
l=[8
,6
3
]
R
5
=[64,255]
R
4
=[32,63]
R
3
=[16,31]
R
2
=[8,15]
R
1
=[0,7]
Fi
gu
re 3.
The
new
p
r
op
ose
d
Type
2
di
vi
si
o
n
o
n
gray
l
e
vel
0
,
255
cont
ai
ni
ng
t
h
e
‘l
o
w
er l
e
vel
’
,
t
h
e
‘m
id
d
l
e lev
e
l’
an
d th
e
‘h
igh
e
r lev
e
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
An
ad
ap
tive st
eg
ano
graph
y sch
e
me ba
sed
on
visu
a
l
qu
a
lity an
d emb
e
d
d
i
n
g
capa
city …
(Mo
jta
ba
B.J.)
57
7
Fi
gu
re
4.
B
l
oc
k
di
ag
ram
of t
h
e i
t
e
rat
i
v
e em
beddi
ng
p
r
ocess
2
2
0
2
255
2
2
0
2
255
(2
)
St
ep
5
:
Calculate the
differe
n
ce
value
s
,
and
,
bet
w
ee
n t
h
e m
i
ddl
e
pi
x
e
l
an
d its t
w
o n
e
i
g
hbo
ri
ng
pi
xel
s
and
o
f
t
h
e co
ve
r-
bl
oc
k
by
E
q
.
(3
).
|
|
,
|
(3
)
St
ep 6
:
Fi
nd the
and
r
a
ng
es
o
f
th
e
r
a
ng
e
d
i
v
i
sion
in qu
estio
n if
and
b
e
lo
ng
t
o
th
e r
a
nges.
St
ep 7:
Obtain the num
b
ers
of
bina
ry secret bits
and
an
d
f
i
nd
t
h
e low
e
r
bo
und
s
and
of
the
corres
ponding
and
ra
n
g
es.
St
ep 8
:
Select
and
b
its of
, an
d tran
sfo
r
m
th
e two
b
it seq
u
en
ces t
o
d
ecimal v
a
lu
es
and
.
St
ep 9
:
C
a
l
c
ul
a
t
e t
h
e ne
w
di
f
f
e
rence
val
u
es
and
usi
n
g E
q
.
(4
).
,
(4
)
St
ep
1
0
:
Calculate
and
v
a
lu
es fo
r th
e fi
rst an
d th
e th
ird p
i
x
e
ls
in th
e
b
l
o
c
k
,
b
y
ob
tain
ing th
e d
i
fferen
ce
b
e
tween
th
e orig
in
al v
a
l
u
e and
th
e n
e
w
v
a
lue of th
em
u
s
ing
Eq
. (5
).
1.
2.
,
1.
2.
(5
)
St
ep 11
:
Detect th
e sen
s
itiv
e
p
i
x
e
ls: fo
ur con
d
ition
s
will be ch
eck
ed
for
in
th
is step
where
1,2
and
(
,
).
I.
If (
0
):
the un
de
r
-flo
w
p
r
o
b
lem
occu
rre
d.
II.
If (
255
):
t
h
e ove
r-
f
l
ow p
r
o
b
l
e
m
happe
ne
d.
II
I.
If (
ϵ
):
t
h
e
di
ffe
r
e
nce
val
u
e
bet
w
een
st
eg
o a
n
d i
t
s
c
ove
r i
s
w
r
o
n
g
.
IV.
If (
0
): the em
be
ddi
ng process
can
be
repeate
d
.
Thu
s
, if at least on
e of th
e
first t
h
ree co
nd
itio
n
s
as well as t
h
e last
co
nd
itio
n
are
satisfied
,
or
0
˅
255
˅
ϵ
˄
0
) is tru
e
, th
e
is sen
s
itiv
e on
t
h
e em
b
e
d
d
e
d
secret b
its and
t
h
e em
beddi
ng
p
r
oce
d
ure ca
n
al
so
be re
peat
e
d
.
St
ep 1
2
:
De
fine two cases
for
wh
er
e
ϵ1,2
and (
,
) as
fo
llo
ws
I.
Case 1
is
sensiti
v
e
Obtain
,
,
,
,
,
Embedding
k
-bits of
into
k
-LSB’s of
and obtain
Calcul
at
e the
dif
f
erenc
e
va
lues
and
between
and
two other
pix
e
ls
of the blo
c
k
Partition
into
som
e
blocks con
t
ain
i
n
g
three
conse
c
utiv
e
Embe
dding
Is the p
i
xe
l
sensitive
?
Compute
Ye
No
Se
le
ct
as
m
i
ddle
pixel of th
e
-blo
ck
Stego-
im
age
’
Cover-
im
age
Find
and
ranges
to which
and
Itera
tio
S
ecret
Dat
a
(
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
57
3
–
58
4
57
8
Find
r
a
ng
e as
th
e pr
ev
iou
s
r
a
n
g
e
thr
oug
h
of the selected di
vision.
R
e
do
em
beddi
ng
p
r
ocess
fr
o
m
st
ep 7 t
o
1
1
fo
r
.
II.
Case 2
is insensit
i
v
e
Consi
d
er the
value of
calculated in
Step10.
When
falls in
to
th
e
u
n
d
e
r-flow
o
r
i
n
t
h
e
o
v
e
r-
fl
o
w
pr
obl
em
, t
h
e sec
o
n
d
val
u
e s
h
oul
d
be c
a
l
c
ul
at
ed i
n
St
ep1
0
.
Fin
a
lly, th
e steg
o-b
l
o
c
k
will
b
e
p
r
o
d
u
c
ed
.
Please n
o
t
e t
h
at th
e abov
e
p
r
oced
ure sho
u
l
d
b
e
rep
eated
for each c
o
ver-bl
ock. Thus,
the secret m
e
s
s
age
will
be em
bedded c
o
mpletely in the cove
r im
age and it
s
steg
o-im
ag
e will b
e
pro
d
u
c
ed
.
3.
3.
Data Extracting Process
For sec
r
et m
e
s
s
age ext
r
action, suppose th
at
t
h
e t
y
pe of
ran
g
e di
vi
si
on a
n
d t
h
e
k
val
u
e use
d
i
n
t
h
e
em
beddi
ng
p
h
a
se are a
v
ai
l
a
b
l
e. M
eanw
h
i
l
e
pl
ease n
o
t
e
t
h
a
t
t
h
ei
r val
u
es
have
bee
n
c
o
n
s
i
d
ere
d
base
d
on
t
h
e
purpose of
hi
ding. This proce
d
ure
is
started with
di
viding t
h
e stego-im
age into
non-overl
appi
ng bloc
ks each
of
whic
h 3×
3 pixels. For each bl
ock as
i
-bl
o
ck
,
k
-
b
i
t
s
of t
h
e m
i
ddl
e
pi
xel
should be selected.
This
sequence
of
bits is placed in t
h
e ri
ghtm
o
st of secret
bits (
). The
n
,
we calculate the diffe
re
nce val
u
es
and
bet
w
ee
n t
h
e m
i
ddl
e
pi
xel
an
d i
t
s
t
w
o
nei
g
h
b
o
r
i
n
g
pi
xel
s
and
of
t
h
e st
e
g
o-
bl
oc
k,
usi
n
g
Eq.
(
6
)
.
|
|
,
|
|
(6
)
The
an
d
r
a
nges to
wh
ich
and
b
e
l
o
ng
ar
e.
Th
e l
o
w
e
r
bo
un
d valu
es
and
and t
h
e
num
ber
of em
bed
d
e
d
bi
t
s
and
are c
hose
n
, conside
r
ing these ra
nges
. T
h
ere
f
ore, t
h
e s
ecret value
s
are
calculated by Eq.
(7).
,
(7
)
In
the
fi
nal ste
p
,
and
values
are c
o
nve
rted i
n
to their
bina
ry sequ
en
ces
b
a
sed on th
e nu
mb
er
o
f
th
e b
its
and
. Th
en
, th
ese
bin
a
ry sequ
en
ces sh
ou
ld b
e
add
e
d to th
e ri
gh
tm
o
s
t o
f
th
e
secret
b
its (
).
Fin
a
lly, th
e secret m
e
ssag
e
wi
ll b
e
ex
t
r
acted
p
r
op
erly withou
t an
y
d
i
stortion
.
3.
4.
Simple Exam
ple of th
e Pr
oposed Method
In th
is section
,
we
will im
p
l
e
m
en
t a sim
p
le ex
am
p
l
e
o
f
t
h
e propo
sed m
e
t
h
od
. Th
e
n
e
x
t
su
bsectio
ns
p
r
esen
t
th
e d
a
ta e
m
b
e
d
d
i
n
g
an
d
ex
tractin
g
p
r
o
cesses. Thu
s
, we
will d
e
m
o
n
s
trate th
at secret b
its are p
r
op
erly
em
bedde
d a
n
d
ext
r
act
ed
usi
n
g
o
u
r
p
r
o
p
o
se
d
m
e
t
hod.
I. D
a
t
a
emb
e
dd
ing p
r
ocess
:
Step1
,
we will u
s
e
Typ
e
1
di
vi
si
o
n
a
n
d
3
for t
h
is e
x
am
ple. T
h
e
secret m
e
ssage is
11010
0001
101
0
and t
h
e se
l
ect
ed bl
oc
k c
ont
ai
n
s
t
h
e t
h
r
ee pi
xel
s
109
and
4
4
and
3
.
Step2
, t
h
e
k-leftmo
s
t
o
f
secret
bi
t
s
sh
oul
d be
re
pl
ac
ed wi
t
h
k-LSB
s
of
. T
h
e
bi
na
ry
val
u
e
of t
h
e m
i
ddl
e pi
xel
i
s
4
4
101100
, so
th
e
d
ecim
a
l v
a
lu
e will b
e
101110
4
6
.
Step3
,
the difference value
is calcu
lated
b
e
tween
4
th
at is th
e d
ecim
a
l v
a
lu
e of
k-LSBs
of
and
6
th
at is th
e deci
m
a
l v
a
lu
e o
f
k-leftmo
s
t
of the secret
bits by
4
6
2
.
Step4
, we
ob
tain
4
6
usi
n
g OP
AP [
21]
.
Step5
, t
h
e
differe
n
ce
val
u
e is calculate
d
betwee
n
an
d
t
h
e
ot
her
t
w
o
pi
xel
s
and
by
|
109
4
6
|
6
3
and
|
3
4
6
|
4
3
.
Step6
,
th
e v
a
lu
e
6
3
is in
and
4
3
is in
consideri
n
g the
Ty
pe
1
d
i
vi
si
on
.
Step7
, the
val
u
es
5
and
5
that are t
h
e
num
b
er of
secret bits a
n
d
3
2
and
3
2
that are the l
o
we
r
bound val
u
es
of t
h
e c
o
rres
ponding
ranges
are
obt
ai
ne
d.
Ste
p
8
, t
h
e
fi
ve
bi
t
s
o
f
t
h
e
sec
r
et
m
e
ssage i
s
s
e
l
ected and c
o
nve
rted into the
decim
a
l value as
10000
1
6
.
In add
itio
n, fi
v
e
of
secret
b
its ar
e
tran
sform
e
d
to
t
h
e
d
ecimal v
a
l
u
e as
11010
2
6
.
Ste
p
9
, t
h
e
ne
w
differe
n
ce
values is
calcul
a
ted
using (4)
as
3
2
1
6
4
8
and
3
2
2
6
4
8
.
Step10
, the
new val
u
es
9
4
and
1
2
are calculated.
Step11
z1
, the
is an insensitive pi
xel
because
of
the (
0
)
an
d
(
255
) and
(
ϵ
)
con
d
ition
s
.
Step12
z1
,
94
sh
oul
d
be
co
nsi
d
e
r
e
d
i
n
acc
or
da
nce
wi
t
h
t
h
e
fi
r
s
t
s
t
age
of
t
h
i
s
st
e
p
.
Ste
p
11
z2
,
b
a
sed
o
n
two
co
nd
itio
ns (
0
) a
n
d
(
0
),
the
is a sen
s
itiv
e p
i
x
e
l.
Step12
z2
, therefore, we
consider
rang
e
o
f
Type
1
di
vi
si
o
n
base
d
o
n
t
h
e sec
o
nd
c
a
se o
f
t
h
i
s
st
e
p
.
The
n
,
t
h
e
e
m
beddi
n
g
process
repeats
step 7 t
o
11
for
.
Step7
, t
h
e
num
ber
of sec
r
et
bi
t
s
an
d t
h
e
l
o
we
r b
o
u
n
d
val
u
e a
r
e
4
and
1
6
base
d o
n
r
a
ng
e of
Ty
pe1
di
vi
si
o
n
.
St
ep8
, t
h
e
fo
ur
b
i
t
s
of t
h
e sec
r
e
t
m
e
ssage i
s
chos
en a
n
d
co
nv
erted in
to th
e
d
ecim
a
l valu
e as
1101
1
.
Ste
p
9
,
th
e
d
i
ff
e
r
en
ce
v
a
lu
e is
c
a
l
cu
la
te
d a
s
16
1
3
29
.
Step10
, th
e new
v
a
lu
e
1
7
is calcu
lated
.
Ste
p
11
, t
h
ere are three con
d
ition
s
(
0
)
and
(
255
) and
(
ϵ
). S
o
,
is an
in
sen
s
itiv
e p
i
x
e
l.
Step12
,
b
a
sed
on
th
e
fi
rst case of th
is
step
, th
e
fin
a
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
An
ad
ap
tive st
eg
ano
graph
y sch
e
me ba
sed
on
visu
a
l
qu
a
lity an
d emb
e
d
d
i
n
g
capa
city …
(Mo
jta
ba
B.J.)
57
9
v
a
lu
e is
1
7
. Fina
lly, the secret message
11010
000
1101
i
s
em
bedded
pr
o
p
e
rl
y
i
n
t
h
e t
h
r
ee
pi
xel
s
9
4
and
4
6
and
1
7
o
f
t
h
e
st
eg
o-
bl
oc
k.
II.
Da
t
a
ex
tracting
pro
cess
:
We use
d
Ty
pe
1
d
i
v
i
sion
and
3
an
d
di
vi
ded
t
h
e c
ove
r i
m
age i
n
t
o
som
e
bl
oc
ks havi
ng
t
h
ree pi
xel
s
i
n
t
h
e em
beddi
ng
ph
ase. At
fi
rst
,
t
h
e val
u
e
of
t
h
e
m
i
ddl
e pi
x
e
l
is
con
v
e
r
t
e
d i
n
t
o
bi
na
ry
val
u
e
as
10111
0
. T
h
e
three LSBs
of
are s
e
lected as the
th
ree righ
tmo
s
t
b
its of secret data. Th
en
, we
c
a
lculate
the differe
n
ce values
|
9
4
4
6
|
4
8
and
|
1
7
4
6
|
2
9
by
Eq
. (
6
). B
a
sed
on
t
h
e
Ty
pe1
di
vi
si
on
a
n
d
t
h
e
ra
nge
t
o
w
h
i
c
h
bel
o
ngs
, t
h
e
val
u
e
s
3
2
and
5
are
obtained. So,
is
4
8
3
2
1
6
by
Eq.
(
7
)
an
d i
t
s bi
nary
val
u
e
i
s
10000
.
M
o
re
ove
r,
4
and
1
6
are consi
d
e
r
ed
using the
Type
1
di
vi
si
o
n
and t
h
e
to
wh
ich
bel
o
ngs
.
B
a
sed o
n
Eq
. (7
),
t
h
e
val
u
e of
2
9
1
6
1
3
is
1101
, whe
n
4
. T
h
e
and
bi
t
sequences
are adde
d to the se
cret
bits
(S).
Finally, we
coul
d e
x
tract
the se
cret
m
e
ssage
1101000
011
01
, correctly.
4.
EX
PER
I
M
E
NTA
L
R
E
SU
LTS AN
D ANALY
S
IS
In th
is sectio
n, we d
e
m
o
n
s
trate th
e effectiv
en
e
ss of our propos
ed
schem
e
com
p
ared with
the Wu
et
al
.’s
[
18]
, t
h
e
Yan
g
et al.’s
[15] a
n
d also t
h
e Khodaei
et al.’s
[
1
9]
m
e
t
hods.
We
pre
s
ent
som
e
expe
ri
m
e
nt
al
resu
lts ob
tain
ed
u
s
ing
10
cov
e
r
im
ag
es
with
512×512
im
age
resol
u
tions. All t
h
e c
o
ver im
ages ha
ve bee
n
trans
f
orm
e
d to grayscale im
ages. Cover im
ages include
Ba
boo
n
,
Ba
r
b
a
r
a
,
Boa
t
,
Camer
a
ma
n
,
Lena
,
Li
vi
ngr
o
o
m
,
Peppe
rs
,
Pirate
,
Tiffa
n
y
and
Zeld
a
. Furt
herm
ore
,
the secret b
its are produced
by a Ra
ndom
Nu
m
b
er
Gen
e
rato
r
(RNG). Gen
e
rally, v
i
su
al q
u
a
lity, h
i
d
i
ng
cap
acity an
d in
fo
rm
atio
n
secu
rity are
u
s
ed
in
eval
uat
i
o
ns o
f
a gi
ven st
e
g
a
n
o
g
r
ap
hy
al
go
ri
t
h
m
.
The pr
op
ose
d
m
e
t
hod p
r
o
v
i
d
e
s
hi
g
h
dat
a
sec
u
ri
t
y
wi
t
h
h
i
gh
er v
i
su
al qu
ality wh
ile its em
b
e
d
d
i
ng
cap
acity will b
e
l
a
rg
er th
an
t
h
e
well-kno
wn
m
e
th
od
s.
The Pea
k
Signal to Noise Ratio (PSNR) val
u
e is used
to
ev
alu
a
te th
e d
i
st
o
r
tion
s
of stego
-
im
ag
e.
W
e
com
pute the PSNR
value in
dB
by Eq. (8),
where t
h
e Mean Square
Er
ro
r
(M
SE) is calc
u
lated, as
sh
o
w
n i
n
Eq
. (9
).
In
Eq.
(
9
)
,
m
is t
h
e s
h
ared size of c
o
ver
and stego
’
im
ages.
1
0
255
(8
)
1
(9
)
Furt
herm
ore,
let
b
e
th
e to
tal b
its
o
f
an emb
e
dd
ed
m
e
ssag
e
i
n
to
a stego-im
ag
e. As sho
w
n
i
n
Eq.
(10), we
calcul
a
te
as t
h
e
ave
r
age c
a
pacity in bit
pe
r
pi
xel
,
w
h
er
e
is
th
e
nu
mb
e
r
o
f
s
e
cr
et b
its
and
is the size
of t
h
e c
ove
r-im
a
ge.
/
(1
0)
4.
1.
Vi
sual
Qu
al
i
t
y
Here, we
will measu
r
e th
e
v
i
su
al qu
ality o
f
steg
o-im
ag
es
p
r
od
u
c
ed
b
y
ou
r
p
r
op
o
s
ed
meth
od
u
s
i
ng
su
bj
ectiv
e an
d
o
b
j
ectiv
e m
ean
s o
f
m
easu
r
emen
t. In
add
itio
n, th
e n
u
m
b
e
r of sen
s
itiv
e p
i
x
e
l
s
will b
e
calcu
lated
to justify the effectivenes
s
of
our
propos
ed
m
e
thod t
o
provide better vis
u
a
l
quality.
4.
1.
1.
Subjec
tive
Me
asureme
nt
Th
e secret messag
e
im
p
e
rcep
tib
ility to
th
e
h
u
m
an
eye is th
e m
a
in
g
o
a
l
o
f
all steg
anog
raph
y
techniques
. In
othe
r words, t
h
e hum
a
n eye, as a
m
ean
s of as a subjective
m
easurem
ent, shoul
d be
unable to
notice
the secret
m
e
ssage
in cove
r. Our firs
t
test
cas
e
m
e
a
s
ure
d
the
stego-im
age by the vision system
. W
e
em
bedde
d t
h
e m
a
xim
u
m
secret
dat
a
usi
n
g t
h
e
Ty
pe1
di
vi
s
i
on a
nd
va
ri
o
u
s
k
val
u
es
o
n
Peppers
c
ove
r-im
age.
Fi
gu
re 5a a
nd
Fi
gu
re 5
b
sh
o
w
t
h
e co
ver a
n
d i
t
s
st
ego. M
o
reo
v
er
, t
h
e di
f
f
e
rence
bet
w
ee
n t
h
e sel
ect
ed regi
o
n
of
t
h
e st
e
g
o
an
d t
h
e
co
rre
sp
o
ndi
ng
re
gi
o
n
o
f
i
t
s
co
ve
r i
s
pr
esent
e
d
i
n
Fi
g
u
r
e
5c.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
57
3
–
58
4
58
0
Fi
gu
re 5.
S
t
eg
o-ima
g
e
qua
lity a
n
a
l
ysis
:
Em
beddi
ng
t
h
e
secre
t
bi
t
s
by
usi
n
g
t
h
e p
r
op
ose
d
m
e
t
hod
wi
t
h
typ
e
1
di
vi
si
o
n
a
n
d
3
,
a) cov
e
r-im
a
ge
,
b) s
t
ego-im
ag
es
,
c) d
i
ffer
e
nc
e b
e
t
w
een th
e s
e
le
cte
d
region
of s
t
ego
-
im
age and
i
t
s
corresponding
cover-image region (
k=3
:
38.95
dB,
806058
bits.
k=4
:
3
6
.
64
dB,
896273
bits.
k=5
:
33.46
dB,
1002440
bits.
k=6
:
32.61
dB,
1157767
bits
Unlike the
Khodaei
et al.'
s
[
19]
t
ech
ni
q
u
e,
ou
r schem
e
pr
event
s
large di
ffe
rence
values
between the
co
v
e
r
an
d its steg
o-
im
ag
e
'
s p
i
x
e
ls. Th
u
s
, the em
er
g
e
n
ce of
sign
if
ican
t
v
i
su
al d
i
st
o
r
tion
is p
r
ev
en
ted
t
h
r
ough
t
h
e p
r
op
ose
d
m
e
t
hod,
an
d t
h
e secret
m
e
ssage i
s
i
nvi
si
bl
e
as t
e
st
ed
by
t
h
e
h
u
m
a
n vi
sual
sy
st
em
(HVS
).
4.
1.
2.
Objec
t
ive Me
asureme
nt
We ex
pe
ri
m
e
n
t
our
p
r
o
p
o
sed
m
e
t
hod
usi
n
g t
h
e
Ty
pe1
and
Type
2
di
vi
si
o
n
s
and
di
f
f
ere
n
t
k
v
a
lu
es
on
th
e cov
e
r im
a
g
es. Tab
l
e 1
presen
ts the num
b
e
r o
f
th
e
sen
s
itiv
e p
i
x
e
ls an
d
t
h
e PSNR
v
a
lu
es
o
f
the steg
o-
im
ages. The
proposed m
e
thod
has
found t
h
e sensitive
pixe
ls
to
pre
v
ent high diffe
re
nc
e
value
s
. As per
a
n
objective m
easurem
ent, the re
sults s
h
o
w
t
h
at
t
h
e PS
NR
val
u
es are
hi
ghe
r
t
h
an
30
(
d
B
)
.
The
n
, i
t
s
em
beddi
ng
p
r
o
cess
was mo
d
i
fied
o
n
t
h
ese sen
s
itiv
e
p
i
xels. Of
cou
r
se,
th
e in
creased
nu
m
b
er o
f
th
e sen
s
itiv
e p
i
x
e
ls
u
s
ing
th
e
Ty
pe2
di
vi
s
i
on i
s
beca
use
of
i
t
s
hi
g
h
e
r
e
m
beddi
ng
ca
pa
ci
t
y
co
m
p
ared
wi
t
h
t
h
e
Ty
pe1
di
vi
si
on
.
I
n
Figu
r
e
6
a
,
w
e
co
m
p
ar
ed th
e PSN
R
v
a
l
u
es of
ou
r pr
opo
sed
m
e
th
o
d
w
ith
t
h
e
W
u
et
al.’s
[1
5]
a
n
d
Yan
g
et al.’s
[
18]
a
nd al
s
o
K
h
o
d
aei
et al.’s
[1
9]
m
e
t
hod
s whe
r
e
t
h
e x-a
x
i
s
prese
n
ts the
stego-im
age and the
y-ax
is sh
ows t
h
e PSNR
v
a
lu
e. As
d
e
m
o
n
s
trated
in
th
is
figure, the m
ean PSNR
valu
e
(
37.65
d
B
)
fo
r th
e steg
o-
i
m
ag
es
pr
odu
ced
b
y
th
e A
d
ap
tiv
e
steg
anogr
aph
y
m
e
th
o
d
u
s
ing
LSB r
e
place
m
e
n
t
an
d PVD
(
w
ith
63
,
3
)
[1
9]
i
s
bet
t
e
r
t
h
a
n
t
h
e c
o
rre
spo
n
d
i
n
g
val
u
es (
3
7.
35
dB
a
n
d
37
.3
3
dB
)
of
t
h
e
st
ega
n
o
g
ra
p
h
y
m
e
t
hod
s
usi
n
g LSB
repl
acem
e
nt
and
P
VD
[1
8]
an
d E
dge a
d
a
p
t
i
v
e L
S
B
m
e
t
hod
(wi
t
h
di
vi
si
o
n
3
-
4)
[1
5]
. H
o
weve
r, t
h
e
quality of ‘Tiffany’ a
n
d ‘Cam
eram
an’
stego-im
ages produc
ed
by the
Ada
p
tive m
e
thod i
s
re
duce
d
due
to the
larg
e
nu
m
b
er o
f
sen
s
itiv
e
p
i
x
e
ls in
t
h
ese i
m
ag
es (calcu
lated
in
Tab
l
e
1
)
. Fu
rth
e
rm
o
r
e, as sh
own
in
Tab
l
e 1,
th
e nu
m
b
er of
sen
s
itiv
e
p
i
x
e
l
s
is lo
w i
n
‘
Zeld
a
’ a
nd al
s
o
i
n
‘
Boat
’ usi
ng
Typ
e
1
. Th
u
s
, the q
u
a
lity o
f
th
em h
a
s
i
n
crease
d
sl
i
g
h
t
l
y
. B
a
sed
o
n
Fi
gu
re
6a,
t
h
e
PSNR
val
u
es
of
o
u
r
m
e
t
hod
(wi
t
h
Ty
pe1
di
vi
si
o
n
,
3
) is
als
o
m
o
re th
an
th
e
Ad
ap
tiv
e steg
an
ograph
y
m
e
th
o
d
. In
add
itio
n, th
e qu
ality o
f
‘Babo
on’ stego
-
im
ag
e p
r
o
duced
b
y
the propose
d
m
e
thod s
o
are
d
in com
p
arison with the othe
rs due to the la
rge
num
ber of sensitive pi
xels in it.
There
f
ore, we
provide a
bette
r
visual
quality
of stego-im
ag
e than these
ot
her m
e
thods.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
An
ad
ap
tive st
eg
ano
graph
y sch
e
me ba
sed
on
visu
a
l
qu
a
lity an
d emb
e
d
d
i
n
g
capa
city …
(Mo
jta
ba
B.J.)
58
1
Tabl
e 1.
Visua
l
qu
a
lity ana
lysis
: Th
e
n
u
m
b
e
r of sen
s
itiv
e
p
i
x
e
ls (
)
and
th
e
P
S
N
R
(
dB
)
val
u
es
of
steg
o-
im
ag
es pr
odu
ced b
y
t
h
e pr
opo
sed m
e
t
h
od
(
w
it
h
Ty
pe
1
an
d
Ty
pe2
di
vi
si
o
n
a
n
d
di
f
f
e
rent
k
val
u
es)
Cover-
im
ages
Sens,
Pix
e
l
PSNR,
dB
Sens,
Pix
e
l
PSNR,
dB
Sens,
Pix
e
l
PSNR,
dB
Sens,
Pix
e
l
PSNR,
dB
Typ
e
1
division
1518
31.
21
960
32.
49
728
35.
88
691
38.
76
Baboon
103
30.
72
99
31.
47
85
34.
18
119
37.
42
Barbara
69
31.
41
29
32.
21
38
35.
27
22
37.
48
Boat
1140
32.
81
419
34.
71
263
37.
94
272
39.
41
C
a
me
r
a
ma
n
26
31.
53
20
32.
61
11
36.
05
12
38.
10
Len
a
1641
32.
61
701
33.
46
377
36.
64
251
38.
95
Pepper
s
386
32.
21
237
33.
24
174
37.
20
154
39.
44
L
i
vingr
oo
m
27
32.
41
16
33.
35
19
37.
44
12
39.
75
Pirate
868
32.
37
820
33.
22
691
37.
04
591
39.
22
T
i
ffa
ny
245
31.
73
49
33.
14
17
37.
05
4
39.
18
Zeld
a
602
31.
90
335
32.
99
240
36.
46
212
38.
76
Average
Typ
e
2
division
4 890
30.
74
4 585
32.
11
3 925
35.
45
3 703
38.
23
Baboon
4 760
30.
29
4 152
31.
06
2 822
33.
66
2 238
37.
01
Barbara
5 390
30.
94
4 520
31.
83
2 654
34.
82
1 945
37.
03
Boat
5 797
32.
20
4 437
34.
34
1 728
37.
55
1 018
39.
00
C
a
me
r
a
ma
n
5 412
31.
13
4 476
32.
18
2 752
35.
68
1 983
37.
64
Len
a
5 541
32.
17
4 510
33.
04
2 840
36.
30
2 143
38.
51
Pepper
s
5 251
31.
73
4 460
32.
77
3 033
36.
84
2 408
38.
98
L
i
vingr
oo
m
5 368
31.
93
4 478
32.
89
2 904
37.
01
2 288
39.
29
Pirate
5 322
31.
88
4 540
32.
76
3 023
36.
53
2 380
38.
87
T
i
ffa
ny
5 571
31.
26
4 493
32.
69
2 602
36.
49
1 769
38.
78
Zeld
a
5 330
31.
42
4 465
32.
56
2 828
36.
03
2 187
38.
33
Average
4.
2.
Embeddin
g
Rate
In
t
h
i
s
t
e
st
ca
se,
we
pr
o
duc
e st
eg
o-i
m
ages by
o
u
r
p
r
op
ose
d
m
e
t
hod
wi
t
h
t
h
e
Ty
pe
1
an
d
Ty
pe
2
di
vi
si
o
n
a
nd
v
a
ri
o
u
s
k
am
ount
s. Ta
bl
e 2 s
h
ows t
h
e ca
paci
t
y
of em
bedde
d secret
bi
t
s
an
d t
h
e a
v
era
g
e c
a
paci
t
y
i
n
bi
t
pe
r pi
xe
l
(
)
f
o
r
eac
h r
a
nge
di
vi
si
o
n
wi
t
h
di
f
f
ere
n
t
k
val
u
es.
As
p
e
r t
h
e
e
xpe
ri
m
e
nt
al
res
u
l
t
s
, t
h
e
capacity and t
h
e
val
u
es usi
ng t
h
e
Ty
pe2
di
vi
si
o
n
are
h
i
ghe
r t
h
a
n
t
h
o
s
e usi
n
g t
h
e
Type
1
di
vi
si
o
n
.
According
t
o
Table 2
,
th
e
values a
r
e i
n
t
h
e [3.037-
4
.
5
29] r
a
ng
e u
s
i
n
g
Type
1
an
d
in
th
e
[
3
.
0
9
7
-
4.667
]
ran
g
e usi
n
g
Ty
pe2
. Also
,
all
avera
g
e values
by
Ty
pe
2
are
hi
g
h
er t
h
a
n
t
h
e
i
r co
rres
p
on
di
n
g
val
u
es
usi
n
g
Ty
pe
1
(
w
he
n
4
and
Ty
pe1
: th
e
av
e
r
ag
e
Cap
a
city
= 912290
bits
,
whe
n
4
and
Ty
pe2
: th
e
av
e
r
ag
e
Capacity
= 935101
bits
).
Th
us, t
h
e
hi
di
n
g
ca
paci
t
y
i
s
i
n
creased
by
usi
n
g
o
u
r
pr
op
ose
d
Ty
pe
2
di
vi
si
on
.
In
Figu
re
6b
,
we co
m
p
are these resu
lts with
th
e
Wu
et al.’s
[18
]
,
Y
a
ng
et al.’s
[
15]
a
n
d K
h
oda
ei
et
al
.’s
[
1
9]
m
e
tho
d
s.
T
h
i
s
fi
g
u
re
sh
o
w
s t
h
at
t
h
e em
bed
d
i
n
g ca
paci
t
y
of
t
h
e p
r
op
ose
d
m
e
t
hod
(
w
i
t
h
Type
2
di
vi
si
o
n
,
3
) i
s
l
a
rge
r
t
h
a
n
t
h
e
St
egan
o
g
ra
p
h
y
m
e
t
hod u
s
i
n
g
LSB
repl
ace
m
e
nt
and P
V
D [
18]
, t
h
e E
d
ge
adapt
i
v
e L
S
B
m
e
t
hod (
w
i
t
h
di
vi
si
on
3-
4) [
15]
an
d al
so t
h
e A
d
apt
i
v
e s
t
egan
og
ra
phy
m
e
t
hod usi
n
g LSB
replacem
ent and PVD
(wit
h
63
,
3
)
[1
9]
.
I
n
Fi
g
u
re
6b
, t
h
e
x
-
a
x
i
s
a
n
d t
h
e y
-
a
x
i
s
s
h
ow
t
h
e
st
ego
-
im
age and the
e
m
beddi
ng ca
pacity. Thus
,
we
provide
a
hi
g
h
e
r em
beddi
ng
capaci
t
y
t
h
an these
othe
r m
e
thods.
4.
3.
Information Security
Here
, t
h
e secur
i
t
y
of t
h
e pro
p
o
se
d
m
e
t
hod i
s
t
e
st
ed i
n
t
e
r
m
s of R
S
and st
e
g
anal
y
s
i
s
det
e
c
t
or at
t
acks.
The RS stega
n
alysis by Fridrich
et al.
[
22]
i
n
2
0
0
1
can s
h
ow e
x
act
l
y
wh
et
her a st
eg
o-i
m
age resi
st
s w
i
t
hout
vi
sual
c
h
eck
.
Thi
s
st
ega
n
al
y
s
i
s
m
e
t
hod cl
a
ssi
fi
es al
l
t
h
e s
t
ego
-
i
m
age pi
xel
s
i
n
t
o
t
h
ree
gr
o
ups
by
usi
n
g d
u
al
st
at
i
s
t
i
cal
m
e
t
hods:
t
h
e
reg
u
l
a
r g
r
ou
p
(
or
),
t
h
e
si
n
g
u
l
a
r gr
ou
p (
or
)
,
and th
e
u
n
u
s
ab
le
g
r
ou
p.
Th
e
relatio
n between
t
h
e
p
e
rcen
tag
e
of the re
gul
ar
gr
ou
p
s
an
d t
h
e si
ng
ul
ar
gr
o
u
p
s
i
s
1
and
1
. Here
and
are p
e
rcen
tag
e
s
with
th
e m
a
sk
, and
and
are p
e
rcen
tag
e
s
with
th
e m
a
sk
–
of t
h
e re
g
u
l
a
r a
nd
t
h
e si
n
gul
a
r
g
r
ou
ps,
res
p
ect
i
v
el
y
.
If
≌
and
≌
, the
stego-
i
m
ag
e will p
a
ss th
e RS attack. Ot
h
e
rwise, t
h
e stego
-
im
ag
e is d
e
tected as a
su
sp
icio
us
o
b
j
e
ct.
Figure 7
a
sho
w
s th
e
resu
lt o
f
th
e RS steg
an
alysis (b
y two
m
a
sk
s
0 1 1 0
and
–
0
1
1
0
)
of
t
h
e
st
eg
o-i
m
ages p
r
od
uce
d
by
t
h
e
p
r
op
o
s
ed
m
e
th
o
d
.
In th
is
figu
re, the x-ax
is presents th
e
em
beddi
ng
ca
paci
t
y
perce
n
t
a
ge a
n
d
t
h
e y
-
axi
s
s
h
o
w
s t
h
e
perc
ent
a
ge
of
t
h
e re
g
u
l
a
r a
n
d t
h
e
si
n
gul
a
r
gr
o
ups
.
Acco
r
d
i
n
g t
o
Fi
gu
re
7a,
we
were
ri
ght
i
n
e
xpect
i
n
g
t
h
at
t
h
e
and
relative
num
bers
are res
p
ectively
equal
t
o
t
h
ose
of
and
i.
e.
≌
and
≌
).
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
JECE Vo
l. 4
,
N
o
. 4
,
Au
gu
st 2
014
:
57
3
–
58
4
58
2
(a
)
(b)
Fi
gu
re
6.
C
o
m
p
ari
s
on
s
bet
w
e
e
n
ou
r
pr
o
p
o
s
e
d
m
e
t
hod a
n
d t
h
e st
e
g
an
o
g
ra
p
h
y
m
e
t
hod
usi
n
g
LSB
re
pl
acem
e
nt
and
P
V
D
,
t
h
e
e
dge
ada
p
t
i
v
e
L
S
B
m
e
t
hod (
di
vi
si
on
3
-
4
)
, an
d th
e ad
ap
ti
v
e
st
eg
anog
r
a
ph
y usin
g LSB
replacem
ent and PVD (
63
,
3
),
a)
Visual qua
lity
co
mparison
: (the p
r
oposed method
with th
e
Ty
pe
1
division and
3
),
b)
Embedding
capacity comparison
: (the proposed method
with
the
Ty
pe
2
div
i
sion and
3
)
Tabl
e 2.
E
m
be
ddi
ng
ca
p
a
ci
t
y
a
nal
ysi
s
: T
h
e
capacity of embedde
d
sec
r
et
bits (
) a
n
d the
avera
g
e
cap
acity in
b
it p
e
r p
i
x
e
l (
)
o
f
steg
o-
im
ag
es pr
odu
ced b
y
t
h
e pr
opo
sed m
e
t
h
od
(
w
it
h
th
e
Typ
e
1
and
Type
2
di
vi
si
o
n
a
n
d
di
ffe
rent
k
val
u
e
s
)
Cover-
im
ages
E,
Bpp
Capacity,
bit
E,
Bpp
Capacity,
bit
E,
bpp
Capacity,
bit
E,
bpp
Capacity,
Bit
Typ
e
1
division
4.
529
1 187 38
0
4.
007
1 050 55
6
3.
626
950 63
7
3.
281
860 31
4
Baboon
4.
529
1 198 69
3
4.
034
1 057 54
2
3.
649
956 60
1
3.
309
867 52
0
Barbara
4.
444
1 165 13
4
3.
860
1 011 89
9
3.
467
908 88
6
3.
126
819 70
3
Boat
4.
408
1 155 61
0
3.
792
994 19
1
3.
415
895 34
6
3.
077
806 76
1
C
a
me
r
a
ma
n
4.
427
1 160 59
2
3.
830
1 004 14
0
3.
426
898 34
7
3.
088
809 61
8
Len
a
4.
416
1 157 76
7
3.
824
1 002 44
0
3.
419
896 27
3
3.
074
806 05
8
Pepper
s
4.
459
1 169 10
6
3.
881
1 017 39
5
3.
483
913 08
9
3.
139
823 06
1
L
i
vingr
oo
m
4.
447
1 165 94
0
3.
858
1 011 60
0
3.
462
907 78
4
3.
120
818 06
1
Pirate
4.
453
1 167 51
8
3.
869
1 014 29
2
3.
470
909 72
9
3.
126
819 66
4
T
i
ffa
ny
4.
399
1 153 35
8
3.
789
993 46
3
3.
380
886 21
7
3.
037
796 39
3
Zeld
a
4.
455
1 168 10
9
3.
874
1 015 75
1
3.
480
912 29
0
3.
138
822 71
5
Average
Typ
e
2
division
4.
622
1 211 67
9
4.
120
1 080 14
8
3.
718
974 68
8
3.
373
884 44
6
Baboon
4.
667
1 223 57
9
4.
151
1 088 38
5
3.
727
977 22
4
3.
368
883 11
2
Barbara
4.
557
1 194 80
1
3.
989
1 045 73
3
3.
550
930 72
7
3.
184
834 82
4
Boat
4.
512
1 182 96
4
3.
941
1 033 27
1
3.
474
910 89
7
3.
097
815 09
6
C
a
me
r
a
ma
n
4.
535
1 188 83
9
3.
971
1 041 08
8
3.
518
922 42
1
3.
152
826 42
9
Len
a
4.
539
1 186 89
8
3.
961
1 038 39
8
3.
513
921 12
5
3.
147
825 03
0
Pepper
s
4.
560
1 195 45
9
4.
014
1 052 43
9
3.
587
937 97
2
3.
213
842 29
5
L
i
vingr
oo
m
4.
547
1 192 16
1
3.
993
1 046 85
6
3.
552
931 33
8
3.
189
836 01
8
Pirate
4.
555
1 194 31
9
4.
003
1 049 52
1
3.
562
933 93
0
3.
198
838 59
0
T
i
ffa
ny
4.
510
1 182 29
7
3.
934
1 031 40
2
3.
474
910 69
6
3.
103
813 66
6
Zeld
a
4.
559
1 195 29
9
4.
008
1 050 72
4
3.
567
935 10
1
3.
204
839 95
0
Average
700000
725000
750000
775000
800000
825000
850000
875000
900000
Capacity, bit
Stego-image
PVD
and LSB method
[18]
Edge adaptive
LS
B
[15]
Adaptive PVD
and
LSB
[19]
34
35
36
37
38
39
40
41
PSNR, dB
Stego-image
PVD
and LSB method
[18]
Edge adaptive
LS
B
[15]
Adaptive PVD and LSB [
19
[
Evaluation Warning : The document was created with Spire.PDF for Python.