TELKOM
NIKA
, Vol.11, No
.3, March 2
0
1
3
, pp. 1691
~1696
ISSN: 2302-4
046
1691
Re
cei
v
ed
De
cem
ber 1
9
, 2012; Re
vi
sed
Jan
uar
y 27, 2
013; Accepte
d
February 1
0
, 2013
Breaking the Digital Video Steganography
Yueqiang Li*
1
, Qiuju Liu
2
1
Departme
n
t of Public T
eachi
ng, Hua
i
h
ua M
edic
a
l Co
lle
ge
Huai
hu
a Cit
y
,
Chin
a, Ph./F
ax: +
86 0745-
238
328
0/23
850
75
2
Departme
n
t of F
o
reign L
a
n
g
u
ages a
nd C
u
ltu
r
e, Huai
hua U
n
iversit
y
Huai
hu
a Cit
y
,
Chin
a, Ph./F
ax: +
86 0745-
286
403
9/
28
549
61
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: li
yue
q
ia
ng@
163.com
*
1
,
ann
elqj @
y
a
h
oo.c
o
m.cn
2
A
b
st
r
a
ct
I
n this paper
w
e
provide a new
digital vide
o stegana
lysis alg
o
rith
m.
W
hen frames of the cover
vide
o ar
e e
m
b
edd
ed w
i
th s
e
cret messag
e
s
,
both n
u
m
b
e
r
of conn
ected
c
o
mpo
nent
an
d
nu
mb
er of
hol
es
w
ill cha
nge
dr
amatica
lly, thu
s
Euler
nu
mb
e
r
of the
stego-f
r
ame w
ill h
a
s
pulsi
ng
incre
a
s
e
. This pro
pos
ed
stegan
alysis a
l
gorith
m
is a
ppl
ied to steg
o-vi
deo, esti
mati
n
g
stegan
ogr
ap
hic al
gorith
m
f
o
r frames c
h
o
s
en
and esti
mating
stegano
gra
phi
c capacity. Experi
m
e
n
tal res
u
lts indic
a
te th
at it
’
s
much si
mp
ler an
d faster
,
mor
e
se
nsitive
and r
o
b
u
st, stegan
ogra
phy r
a
te
as s
m
a
ll as
0
.
0154
% ca
n b
e
reli
ably
detect
ed.
It
’
s
effective
and efficie
n
t to detect any format video, b
e
caus
e this
algorith
m
is onl
y correlated w
i
th the connec
te
d
compo
nents
nu
mb
er
and
h
o
l
e
s number of th
e frames.
Key
w
ords
:
st
egan
ograph
y,
stegan
alysi
s
, digital vide
o, Euler num
be
r
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 fast developmen
t of multimedia in
formati
on techn
o
log
y
, as well a
s
cyber
techn
o
logy, informatio
n secu
rity has g
a
ined mo
re a
ttention than ever. Since t
he ea
rly 199
0s,
informatio
n h
i
ding ha
s em
erge
d as
an
increa
singly
active re
sea
r
ch a
r
ea. The
r
e ha
s be
en
an
increa
sed int
e
re
st both in
digital wate
rmark for
the
purp
o
se of in
tellectual p
r
o
perty prote
c
ti
on,
and in ste
gan
ogra
phy for the purpo
se of
covert comm
unication.
With the ad
vent of steganog
rap
h
ic t
e
ch
ni
qu
es, there h
a
ve been fatal threats to
cyberse
cu
rity. Thus in t
he fiel
ds
of cyberse
cu
rit
y
, there exists a
n
incre
a
se
d interest
in
stega
nalysi
s
,
i.e. the sci
ence of
iden
tifying stega
nographi
c se
cret me
ssag
e. In the future
stega
nog
rap
h
y
vs steganal
ysis will em
erge as the fo
cus of cybe
rsp
here.
There are two
stega
nal
ysis a
pproa
ches,
in
cludi
n
g
qualitative
stega
nalysi
s
[1] and
quantitative stega
nalysi
s
[
2
]. Qua
litative stegan
al
ysis aim
s
to detect th
e existen
c
e
o
f
stega
nog
rap
h
i
c secret m
e
ssag
e, whi
l
e qua
nt
itative stega
nal
ysis b
a
sed
on qu
alita
t
ive
stega
nalysi
s
aims to estimate the capacity[
3], to
decode the se
cret key[4]
, to
guess the
stega
nog
rap
h
i
c algorith
m
Error! Re
fer
e
nce sourc
e
not found.
, u
l
timately, to
extract the hidden
information.
Steganalysi
s
aims to
di
scover the hi
dden i
n
form
ation from
a
given
cover media.
Ho
wever, it’s
quite difficult. There a
r
e th
ree mai
n
chal
lenge
s. Firstly, there exist wide availa
bil
i
ty
of covers, su
ch a
s
digital
video, image,
audio, text
etc. Seco
ndly
,
there exist l
a
rge
numb
e
rs of
covers. Thi
r
dl
y, there exist
wide
ran
ge of
algorith
m
s.
T
hus it’
s
rath
er difficult to de
tect the hid
d
e
n
informatio
n from the
co
vers. Th
eref
ore the
pr
i
m
ary go
al of stega
naly
s
is i
s
qu
alitative
stega
nalysi
s
.
Although th
e hidde
n info
rmation
ca
n
n
o
t be retri
e
ved by qualita
t
ive steganal
ysis
itself, once the existence
of t
he steganog
rap
h
ic d
a
ta is discov
e
red, the pu
rpo
s
e of covert
comm
uni
cati
on ca
n be pre
v
ented via de
letion and a
c
t
i
ve attack, etc.
2. Digital Video Stega
n
al
y
s
is
2.1 Euler Nu
mber
Euler num
be
r of image
(frame
) is o
ne of
the most impo
rt
ant cha
r
a
c
te
ristics in
topologi
cal, i
t
remain
s i
n
variant u
n
d
e
r tran
sl
atio
n, rotation, scaling, an
d
rubb
er
she
e
t
transfo
rmatio
n of the image.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 3, March 2
013 : 1691 –
1696
1692
The Euler n
u
m
ber i
s
defin
ed as Equ
a
tio
n
1 belo
w
[6]:
E
=
C
-
H
(
1
)
Whe
r
e C i
s
numbe
r of conne
cted
co
mpone
nts.
H is numbe
r o
f
holes. Hol
e
s are the
backg
rou
nd region b
a
sed
on the foreg
r
ound of the b
o
rde
r
. E is Euler num
ber.
2.2 Qualita
t
iv
e
Steganaly
s
is
The di
gital video i
s
comp
ose
d
of a
serie
s
of
still image
s, the
s
e imag
es
are call
ed
frame
s
. Usua
lly sce
ne
s in
a sh
ot ch
ang
e little,
thus the Eule
r nu
mber
of the frame
s
cha
n
g
e
s
indistinctively.
Once digital v
i
deo fra
m
e
s
a
r
e em
bedd
ed
with se
cret i
n
formatio
n, u
s
ually the d
a
ta of the
frame
s
will be
modified,
C and H
will ch
ange distin
ct
i
v
ely, Thus Eu
ler nu
mbe
r
of
the steg
o-fra
m
e
will has pul
si
ng increase. Therefore thi
s
approac
h can be exploit
ed to detect the existence
of
stega
nog
rap
h
i
c se
cret messag
e.
After digital
video fram
es are e
m
be
dd
ed with
hidd
en informatio
n, in ord
e
r t
o
dete
c
t
cha
nge
s of the Euler nu
mber, we
ch
oose 2 vi
deo
sample
s. Sample 1, 500 frame
s
, 360
×288
pixels, 25fp
s
,
AVI, none vi
deo
com
p
re
ssion, true
col
our. Sam
p
le
2, 372 f
r
ame
s
, 720
×5
76
pi
xels,
25fps, AVI, DX50 video co
mpre
ssion, true col
our.
Secret inform
ation is re
sp
ectively embedde
d
in the
frame
s
of 10th, 20th, 30th ,…,
the
stega
nog
rap
h
y
rate of frames is 1
0
%.
Cover medi
a
is em
bed
de
d with 4
0
×
40, 32
× 3
2
, 24 × 24, 1
6
× 16,
8 ×
8
pixels
mono
ch
rome BMP
images respe
c
tively.
Steganog
ra
p
h
ic alg
o
rithm
is as follo
ws:
To tran
sform
su
ccessive 8 × 8 pixel blocks
of the fra
m
e, the secret message h
a
s bee
n
embed
ded in
to intermedi
a
t
e frequen
cy of the DCT
coefficient, thu
s
re
sulting in
5 stego-vid
e
o
s
for each sa
m
p
le.
To dete
c
t re
spe
c
tively the 2
sampl
e
s Euler n
u
mb
er a
nd the
1
0
steg
o-vide
os Eul
e
r
numbe
r, re
su
lts are
as
sh
own i
n
Fig. 1
,
Fig. 2,
Fig. 3, Fig. 4. To
observe
cle
a
rly and vivid
l
y,
results of fra
m
es fro
m
1st
to 100th are
shown.
Figure 1. Sample 1. cove
r-obje
c
t Euler
numbe
r
Figure 2. Sample 1. 40
×
40 steg
o-o
b
je
ct Euler
numbe
r (steg
anog
rap
h
y ra
te 1.54%)
Figure 3. Sample 2. cove
r-obje
c
t Euler
numbe
r
Figure 4. Sample 2. 40
×
40 steg
o-o
b
je
ct Euler
numbe
r (steg
anog
rap
h
y ra
te 0.0154%)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Title of m
anuscript is
sho
r
t and cle
a
r, im
plies resea
r
ch results (First Author)
1693
As for Sampl
e
1, embedd
ed with 40 × 40, cover-obj
ect Euler nu
mber an
d ste
go-o
b
je
ct
Euler n
u
mbe
r
are d
e
mo
nst
r
ated
re
spe
c
t
i
vely in
Fig. 1 and Fi
g. 2
,
stegan
ogra
phy embe
ddi
ng
rate is 1.5
4
%
.
As for Sam
p
le 2, embe
d
ded with
8
×
8, cover-obj
e
c
t Euler n
u
m
ber a
nd ste
g
o
-
obje
c
t Euler numbe
r are
demon
strat
ed re
sp
ecti
v
e
ly in Fig. 3 and Fig.
4, stegan
og
raph
y
embed
ding
ra
te is 0.0154%
.
As sh
own in
Fig. 2, Fig. 4,
once so
me frames
are
em
bedd
ed wi
th
se
cret info
rm
ation, we
can
comp
are
the cover-fra
m
e Euler nu
mber
wi
th the stego-f
r
ame
Euler num
ber, the stego-fram
e
Euler num
be
r will ha
s pu
lsing in
crea
se, the incre
a
s
e pa
ce is 5
to 400. Ad
ditionally, more
embed
ded, la
rge
r
is the pa
ce.
Pulsing in
cre
a
se of Eule
r numbe
r is ev
aluated by Eq
uation 2 bel
o
w
D
=
E
i
-
(
E
i
-
1
+
E
i
+
1
)
/
2
(
2
)
Whe
r
e Ei is
Euler numb
e
r of Frame i,
Ei-1 is Euler numbe
r of Frame i-1, Ei+1 is Euler
numbe
r of Frame i+1, D i
s
differen
c
e.
If D is greate
r
than or equ
al
to limit
，
then Frame i is e
m
bedd
ed wit
h
covert info
rmation.
To dete
c
t re
spectively the
5 steg
o-vide
os of Sam
p
le
1
，
utilizi
ng l
i
mits 10, 20,
30, 40
and 50.
To evaluate the re
sult, this paper a
dopt
s the following
stand
ard
s
:
Dete
ctable ra
te P1 is given by Equation 3 belo
w
P
1
=
M
1
/
M
(
3
)
Whe
r
e M1 i
s
detecte
d steg
o-fram
e, M is embed
ded
stego-f
r
ame
Missi
ng repo
rt rate P2 is given by Equation 4 belo
w
P
2
=
M
2
/
M
(
4
)
Whe
r
e M2 i
s
detecte
d non
-stego
-fram
e
, M is embe
dd
ed steg
o-fra
m
e.
False al
arm p
r
oba
bility P3 is given by Equation 5 bel
o
w
P
3
=
N
1
/
N
(
5
)
Whe
r
e N1 is
detecte
d non
-stego
-fram
e
, N is no
n-steg
o-fram
e.
As sh
own in
Table 1, De
tectable
rate
P1max= 9
4
%
, Missing
report
rate P2
min=6%,
False al
arm probability P3min=0.89%.
Take th
e thre
e rate
s into consi
deration,
the
app
rop
r
ia
te limit is 20 to 40. Limit in
cre
a
ses
with the ca
pa
city of the cover-f
rame.
2.3 Estimati
ng Stegan
og
raphic Algor
ithm for Fra
m
e Chose
n
The dist
ributi
on of the co
vert informati
on ca
n be e
a
sily judg
ed
from the figu
re of the
sampl
e
Euler number. Th
us it’s po
ssi
b
l
e to estimate the stegan
ogra
phi
c alg
o
rithm for fra
m
e
c
h
os
en
.
The dista
n
ce betwee
n
the stego-frame
s
is con
s
tant, which is
call
ed fixed
stega
nog
rap
h
i
c algo
rithm for fram
e cho
s
en.
The distan
ce between
the stego-f
r
a
m
es is
vari
able, which
is called non-fixed
stega
nog
rap
h
i
c algorith
m
for frame ch
ose
n
, se
le
cted by some
eigenvalue
s, for instance,
indep
ende
nt comp
one
nt a
nalysi
s
(ICA
)
[7]
，
motion vec
t
or[8],
skewness
[
9], kurtos
is
[10],Euler
numbe
r etc.
2.4 Estimati
ng Stegan
og
raphic Capa
cit
y
As shown in
Fig. 2, Fig.
4, frame
s
a
r
e embe
dde
d
with differen
t
capa
citie
s
of cove
r
t
informatio
n, Euler num
be
r incre
a
ses
with the cap
a
citi
es.
To evaluate
and cal
c
ula
t
e non-stego
-frame
Eul
e
r numbe
r an
d stego
-fram
e
Euler
numbe
r of t
he 5
s
tego
-video
s of Sa
mple 2
on
a
v
erage
re
sp
e
c
tively. Resu
lts are sho
w
n in
Table 2.
As sho
w
n in Table 2., Euler numbe
r incre
a
ses
with the capa
citie
s
of covert informatio
n
embed
ded i
n
frame
s
. Furt
herm
o
re, Eul
e
r nu
mbe
r
in
cre
a
ses i
n
proportio
n
to th
e ca
pa
cities
of
covert info
rm
ation. i.e. Euler num
ber in
crea
se
s
app
ro
ximately linearly as capa
cities in
cre
a
se.
The expe
rim
ents
sho
w
th
at the increa
se of Eule
r n
u
mbe
r
is
clo
s
ely related to
not only
the ca
pa
city but also the
distrib
u
tion of
the se
cret informatio
n.
If it’s evenly d
i
stribute
d
in t
h
e
frame
s
, then Euler num
ber incre
a
ses di
stinctively
on
average. If it’s unevenly d
i
stribute
d
in the
frame
s
, then Euler num
be
r incre
a
ses in
distin
ctively on averag
e.
Relatio
n
bet
wee
n
e
s
timat
i
ng capa
city
and in
crea
se
of avera
ge
Euler n
u
mb
er is
sho
w
n
in Table 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 3, March 2
013 : 1691 –
1696
1694
3 Simulation Results
Utilizing the proposed
approach, we make a
sim
u
la
tion detection based on the steo-
video of Sample 3. Sample 3, 720×576
pixels,2
8
2
frames, 25fps, AVI, DX50 vide
o comp
re
ssi
o
n
,
true c
o
lour.
Dete
ction me
thods a
r
e a
s
follows.
Step 1 : Qual
itative stegan
alysis. As
sh
own i
n
Fi
g. 5
is Sampl
e
3.
Euler n
u
mbe
r
. When
ob
serve
Fig. 5 ca
refu
lly, we find that Euler nu
mber in
crea
ses sha
r
ply in
some frame
s
. To be mo
re
obje
c
tive, Equation 2 i
s
used a
s
scree
n
i
ng conditio
n
. Due to
the la
rge capa
city o
f
Sample 3
(7
20
×
576 pixels), if D
≥
40, then 37 frames are qua
lified. Ther
ef
ore the re
sul
t
is that secret
informatio
n is embedd
ed in
Sample 3.
Table 1. Re
sults of the 5 stego-vide
os o
f
Sample 1.
Capacit
y Limit
P1
P2
P3
40×40
D
≥
10
D
≥
20
D
≥
30
D
≥
40
D
≥
50
86.00%
84.00%
82.00%
68.00%
54.00%
86.00%
48.00%
22.00%
16.00%
8.00%
9.56%
5.33%
2.44%
1.78%
0.89%
32×32
D
≥
10
D
≥
20
D
≥
30
D
≥
40
D
≥
50
88.00%
86.00%
74.00%
54.00%
38.00%
140.00%
50.00%
22.00%
20.00%
8.00%
16.00%
5.56%
2.44%
2.22%
0.89%
24×24
D
≥
10
D
≥
20
D
≥
30
D
≥
40
D
≥
50
94.00%
86.00%
70.00%
52.00%
32.00%
136.00%
44.00%
16.00%
14.00%
6.00%
15.11%
4.89%
1.78%
1.56%
0.67%
16×16
D
≥
10
D
≥
20
D
≥
30
D
≥
40
D
≥
50
94.00%
80.00%
66.00%
46.00%
20.00%
136.00%
42.00%
16.00%
14.00%
8.00%
15.11%
4.67%
1.78%
1.56%
0.89%
8×8
D
≥
10
D
≥
20
D
≥
30
D
≥
40
D
≥
50
86.00%
70.00%
50.00%
20.00%
14.00%
136.00%
44.00%
16.00%
14.00%
8.00%
15.11%
4.89%
1.78%
1.56%
0.89%
Table 2. Ca
p
a
citie
s
& effects on Eule
r n
u
mbe
r
Table 3. Ca
p
a
citie
s
& effects on Eule
r n
u
mbe
r
Table 4. Re
sults of Sampl
e
3
Capacities
A
verage
Euler number
of non-stego
-
frame
A
verage
Euler number
of stego-
frame
Increased
A
verage
Euler number
of stego-
frame
40×40 143.07
244.95
101.88
32×32 143.07
218.35
75.28
24×24 143.07
201.19
58.12
16×16 143.07
189.95
46.88
8×8 143.07
157.27
14.20
Capacities
(bit)
Increase of
averag
e Euler
nu
m
b
e
r
E
v
en distr
i
bution
Uneven distr
i
butio
n
1000~
200
0
80~12
0
180~2
3
0
500~1
000
50~80
130~1
8
0
200~5
0
0
30~50
80~13
0
50~20
0
10~30
30~80
P1 P2
P3
92.
11%
2.
63%
0.
40%
Step 2 : Estimating ste
g
a
nographi
c al
gorithm fo
r
frame cho
s
en.
Acco
rdin
g to Fig. 5 and
the
scree
n
ing
re
sults, we find o
u
t that Sample 3 has
two p
r
ope
rtie
s. Secret info
rmati
on is emb
edd
ed
in every other frame in so
me shot
s. Therefo
r
e the
estimating al
gorithm for frame ch
osen is
based on even / odd frames. While in so
me other
sh
o
t
s, nothing is embed
ded in.
Therefore the
estimating al
gorithm for f
r
ame
cho
s
e
n
is based on
eigenvalue
s of frames a
nd even / odd
frames
.
Step 3 : Estimating stega
nographi
c ca
pacity. In Sa
mple 3, the 37 stego-f
r
am
es Euler num
ber i
s
382.51,
while
the no
n-steg
o-fram
es Eul
e
r n
u
mbe
r
i
s
181.17, the
Euler
numb
e
r i
s
in
crea
sed
b
y
201.34 o
n
averag
e. According to Table
3, the capa
cit
y
is 500 to 20
00 bit.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Title of m
anuscript is
sho
r
t and cle
a
r, im
plies resea
r
ch results (First Author)
1695
In fact, in
Sa
mple 3 there
are 38 steg
o-fram
es in 282 cove
r-fra
mes, each embede
d
cap
a
cit
y
is 32
×
32bit.
Figure 4. Sample 3.Euler
numbe
r
Steganog
ra
p
h
ic alg
o
rithm
is as follo
ws:
Do n
o
t nam
e the ste
g
o
-
frame
s
, but
to sele
ct the
qualified fra
m
es
who
s
e
kurto
s
i
s
eigenvalu
e
is greate
r
than
20 a
s
well
as
even num
ber frame
s
. As for V compo
n
ent of HSV colo
r
spa
c
e, to tra
n
sform su
cce
ssive 8
× 8 p
i
xel bl
ocks of
the frame, the se
cret me
ssage h
a
s b
e
e
n
embed
ded int
o
interme
d
iat
e
freque
ncy o
f
the DCT co
efficient, each is embe
dde
d with 1 bit.
P1,P2 and P3 are sho
w
n i
n
Table 4.
As sho
w
n in
Table 4, Steg
analysi
s
re
sul
t
is reliable.
4. Conclusio
n
Euler nu
mbe
r
is se
nsitive to image d
a
ta, onc
e th
e fra
m
es i
s
modifi
ed, Euler n
u
m
ber
will
has pul
sin
g
increa
se, it ca
n be applie
d to video and image ste
gan
alysis. It’s much
simple
r a
nd
faster,
m
o
re
sen
s
itive
an
d
rob
u
st, steg
anog
rap
h
y
ra
te
as sm
all as
0.0
154% can be relia
b
l
y
detecte
d. Euler num
ber h
a
s topol
ogi
ca
l characte
ri
st
ics, it remai
n
s invaria
n
t under tran
slati
on,
rotation, scal
ing, and ru
b
ber sheet transfo
rmat
io
n
of the image. Therefore, this prop
o
s
ed
stega
nalysi
s
algorith
m
is robu
st. This a
l
gorithm
i
s
effective and e
fficient to detect any form
at
video, becau
se it is only correlated
wi
th
the Euler nu
mber of the frames.
Ackn
o
w
l
e
dg
ment
The wo
rk on
this pager is sup
porte
d by
Hunan Provinci
al Science & Technolo
g
y
Dep
a
rtme
nt of China(No. 2
012GK3
033
).
Referen
ces
[1]
Andre
w
D. Ker.
F
ourth-order structural steg
ana
lysis and a
nalysis of cove
r assumpti
ons
.
In Security
,
Steganogr
aphy and Water-marking
of Multim
edia Cont
ents
VIII. 2006; 6072: 25–38.
[2]
Jan Kodovsk
y
,
Jessica Fridrich.
Quantitativ
e stega
nalys
is
of LSB embe
ddi
ng i
n
JPEG do
mai
n
. in
Proc.12th ACM
W
o
rkshop on
Multimed
ia
a
n
d
Securit
y
. R
o
m
a
. 2010: 1
87–
1
98.
[3]
F
a
rzin Yag
h
m
aee
and M
ans
our Jamza
d
. Estimating
w
a
t
e
rmarkin
g ca
p
a
cit
y
in gr
a
y
s
c
ale im
ag
es
base
d
on im
ag
e compl
e
xit
y
.
EURASIP Jour
nal o
n
Adva
nc
es in Sig
nal Pr
ocessi
ng.
20
10
:
1–9.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NIKA
Vol. 11, No. 3, March 2
013 : 1691 –
1696
1696
[4]
J. F
r
idrich, M.
Golja
n and D. Souka
l
.
Search
ing for the steg
o key.
In. Proc. SPIE, Electronic Imagin
g
,
Securit
y
, Ste
g
a
nogr
aph
y, a
nd
W
a
termarkin
g of
Multimed
ia
Conte
n
ts VI. 2
004; 53
06:7
0
-8
2.
[5]
J. F
r
idrich, M.
Golja
n, and D. Hog
ea.
Stega
n
a
lysis of JPEG ima
ges
: break
ing the F
5
alg
o
r
ithm.
In: F.
A. P. Petitcolas (ed.), Proc. o
f
Information Hidin
g,
5th Int.
W
o
rkshop, IH
200
2, Noord
w
i
j
k
erho
ut,
T
h
e
Nether
lan
d
s, L
NCS 20
03;2
5
7
8
:310-
32
3.
[6]
X
.
Lin, Y. Wu, D. Fu, W.
Qian. A
nal
ysis
of co
mputatio
n com
p
le
xit
y
o
n
Eu
le
r Number
of bi
nar
y ima
ge.
Journ
a
l of Micr
oel
ectronics &
Co
mp
uter
. 200
7; 25(7):14-
16.
[7]
S. Bounkon
g, B.
T
o
ch, D.
Saad, and D. L
o
w
e
. ICA for
w
a
t
e
rmarki
ng
digit
a
l imag
es.
Journal of
Machi
ne Le
arn
i
ng R
e
searc
h
. 200
2; Vol.1:1–
25.
[8]
J. Z
hang, J. Li, and L. Z
han
g.
Video w
a
ter
m
ark techni
qu
e i
n
moti
on vect
o
r
.
In Proc. 14th Computer
Graphics a
nd Image Proc
essi
ng. Brazil
ian S
y
mp. 2
001: 1
7
9–1
82.
[9]
Z
H
OU Z
h
i-
yua
n
, LI Yue-q
i
an
g, SUN
Xin
g
-
m
i
ng. Vi
deo
watermarki
ng a
l
gorithm
base
d
on ske
w
n
ess.
Journ
a
l of Co
mputer Eng
i
n
eeri
ng an
d App
lica
t
ion.
200
8; 44(
25):96-
19
5.
[10]
LI Yue-q
i
a
ng,
QUAN T
ong-g
u
i,
Z
H
OU Z
h
i-
yuan, SU
N
Xi
ng
-ming. Ap
plic
at
ion of
kurtosis eig
enva
l
u
e
i
n
vide
o w
a
terma
rking.
Jour
nal
of Computer E
ngi
neer
in
g and
Applic
ations
. 2
009; 45(
14):9
0
-
92.
Evaluation Warning : The document was created with Spire.PDF for Python.