TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 6386 ~ 6392
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.573
7
6386
Re
cei
v
ed Fe
brua
ry 3, 201
4; Revi
se
d
April 6, 2014;
Acce
pted April 25, 2014
Image P
r
otection by Intersectin
g Signatures
Chun
-Hung Chen
1
, Yuan
-Liang Ta
ng*
2
, Wen-Sh
y
o
ng Hsieh
1,3
, Min-Shiang
H
w
a
n
g
4
1
Departme
n
t of Computer Sci
ence a
nd En
gi
neer
ing, Nati
on
al Sun Yat-se
n
Universit
y
2
Departme
n
t of Information M
ana
geme
n
t, Chao
ya
n
g
Univ
e
r
sit
y
of T
e
chnol
og
y
3
Departme
n
t of Computer Sci
ence a
nd Infor
m
ation En
gi
ne
erin
g, Shu-T
e
Univers
i
t
y
4
Departme
n
t of Computer Sci
ence a
nd Info
r
m
ation En
gi
ne
erin
g, Asia Uni
v
ersit
y
*Corres
p
o
ndi
n
g
author, e-ma
i
l
:
y
l
ta
ng@c
y
ut.
edu.t
w
A
b
st
r
a
ct
In this p
a
p
e
r, w
e
prop
ose
a
n
exact
i
m
ag
e
auth
enticati
o
n
sche
m
e
that
c
an, in
the
be
st case,
detect i
m
a
ge t
a
mperi
ng w
i
th
the acc
u
racy o
f
one
pixe
l.
Thi
s
method
is
ba
sed o
n
co
nstru
c
ting b
l
ocks
in
th
e
imag
e i
n
s
u
ch
a
ma
nner
that
they i
n
tersect
w
i
th one
a
noth
e
r i
n
d
i
fferent
directi
ons. Suc
h
a
tech
niq
u
e
i
s
very useful to i
dentify w
hether
an ind
i
vid
u
a
l
i
m
a
ge p
i
xel h
a
s
been ta
mp
ere
d
w
i
th.
Ke
y
w
ords
: inf
o
rmation security
, digital w
a
te
rmark
i
ng, i
m
a
g
e
authe
nticati
o
n, digita
l sig
nat
ures
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
As digital technolo
g
ies a
d
v
ance, mo
re and more pu
blicatio
ns a
r
e
produ
ce
d in digital
formats a
nd tran
smitted via the Internet
. A
ccomp
anyi
ng su
ch adva
n
ce, ho
weve
r, unauthori
z
e
d
use, ill
egal
copyin
g, an
d
malici
o
u
s
modificati
o
n
of
digital produ
cts have
be
come
se
riou
s
probl
em
s. Re
sea
r
che
r
s th
u
s
try to find v
a
riou
s
ways to protect di
gital pro
d
u
c
ts;
solution
s in
clu
de
copyri
ght a
ssertion,
conte
n
t
authent
ication, etc. In th
e area of im
a
ge content
au
thenticatio
n, the
integrity of a
n
image
is
regarded ve
ry
importa
nt a
nd mu
st therefore
be real
ized. A
com
m
on
approa
ch i
s
the u
s
e
of d
i
gital waterm
arki
ng te
ch
ni
que
s. Di
gital
wate
rma
r
ki
n
g
serve
s
m
a
ny
purp
o
ses, for
example, pro
o
f of ownersh
ip, cont
ent au
thenticatio
n, copy
control,
and so on.
Re
sea
r
che
r
s
have devel
op
ed vari
ou
s i
m
age
authe
n
t
ication te
chn
i
que
s to d
e
te
ct if an
image
ha
s e
x
perien
c
e
d
u
nautho
rized
modificatio
n
.
Some of th
e
m
ca
n o
n
ly d
e
tect
wheth
e
r the
image a
s
a whol
e ha
s b
een altered. Others may
have the add
itional ca
pabi
lity to detect
if a
certai
n pa
rt of the image
has be
en ta
mpered with.
Liu
et al.
[1] studied the
Zenike mom
ent
values which are ge
nerate
d
from low DWT su
bba
nd
s. They found
that the quantized value
s
are
robu
st to
co
mmon p
r
o
c
e
ssi
ng o
perations
but fra
g
i
l
e to mali
cio
u
s atta
cks.
Therefore, th
ey
embed
ded th
e wate
rma
r
k by quantizi
n
g the Zernike moment v
a
lue
s
, and th
e locatio
n
s
(i
.e.,
blocks) suffe
red
f
r
om ma
liciou
s
attacks can
be
id
entified throu
gh exami
n
in
g the
extract
ed
values.
Thei
r method
ha
s mode
rate
robu
stne
ss a
gain
s
t JPEG
co
mpressio
n. In Rawat
and
Rama
n’s
sch
e
me [2], two
cha
o
tic ma
ps a
r
e u
s
e
d
in ord
e
r to
enha
nce the
se
curity of
the
watermarke
d
image
s. The
pixels in th
e i
m
age
are
disturbed
u
s
ing
the first
cha
o
tic ma
p an
d a
r
e
further sepa
rated into
bit
plane
s
with t
he le
ast
signi
ficant bit
use
d
for
wate
rm
ark em
bed
din
g
. A
binary
wate
rmark i
s
scra
m
bled
by the
se
con
d
c
haot
ic
ma
p.
The
watermarke
d image
s can a
v
oid
cou
n
terfeiting
attacks. Xi’
a
n [3] scrambl
ed a
bi-l
evel
watermark
by the Arnold
transfo
rm, a
n
d
the
Huma
n Visua
l
System is u
s
ed
to dete
r
mine the
qua
ntization
step
. The
scram
b
led waterm
ark i
s
then in
se
rted
into the l
o
w
DWT coeffici
ents. Ta
mpe
r
are
a
s can th
en b
e
lo
cali
zed by
com
p
a
r
ing
the extra
c
ted
and th
e o
r
ig
inal waterm
a
r
ks. Patra
et
al.
[4] co
nvert the i
m
ag
es i
n
to the
DCT
domain a
nd q
uantize the lo
w-frequ
en
cy coeffici
ents a
c
cordi
ng to th
e target level
s
dete
r
mine
d by
the Chi
n
e
s
e
Remai
nde
r T
heorem. The
i
r metho
d
is com
putation
a
lly efficient
and i
s
abl
e
to
withsta
nd
su
ch atta
cks a
s
JPEG comp
re
ssio
n, sh
arpenin
g
, and
brighte
n
ing.
Qi
et al.
[5] us
ed
two co
ntent-based watermarks to p
r
o
t
ect the im
ag
es. One of t
hem is g
ene
rated by an e
dge
detecto
r for
the purpo
se
of detectin
g
tiny c
han
ge
s, and th
e
other i
s
ge
n
e
rated f
r
om
the
relation
shi
p
b
e
twee
n the
wavelet coeffic
i
ents
for localiz
i
ng
tampered regions
.
Both watermarks
are
emb
edd
ed into
mid
d
le- and
hig
h
-fre
que
ncy
DWT
coeffici
ents. Fi
nally, the g
ene
rat
e
d
watermarks a
nd extra
c
ted
watermarks a
r
e co
mpa
r
ed
to authenticate the image,
and a mali
cio
u
s
attack i
s
iden
tified if error
pixels are clu
s
tere
d togeth
e
r. Their m
e
thod is robu
st
against seve
ral
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age Protection by Intersecting Sign
ature
s
(Chu
n-Hung Chen
)
6387
image p
r
o
c
e
ssi
ng op
erati
ons, in
cludin
g
JPEG com
p
re
ssi
on. In Wu’
s
wo
rk [
6
], the image is
divided into
b
l
ocks, a
nd all
hashe
s de
ri
ved from
the
MSBs of ea
ch bl
ock a
r
e
further
en
cod
e
d
usin
g a
n
e
r
ro
r
corre
c
ting
code
(ECC).
T
he p
a
riti
e
s
, rather than
th
e code
wo
rd
s, of the
ECC
are
sep
a
rate
d an
d embe
dded
into the LSBs of eac
h
block. Du
rin
g
authentic
ation, the origi
nal
hashe
s can b
e
re
cove
red if
the num
ber
of tamper
ed
blocks is l
e
ss than a th
re
sh
old. The
hash
of
a blo
c
k is p
r
o
duced
and
co
mpared
with t
he o
r
igin
al to
identity if the
block i
s
tam
p
ered
with.
Thi
s
method ha
s fine gra
nula
r
ity on detectin
g
tampered re
g
i
ons.
Although
ma
ny tech
niqu
e
s
h
a
ve
been
prop
osed
for
image
authe
n
t
ication, m
o
st
of them
can
only det
ect if an im
a
ge o
r
pa
rt of
it, as a
whol
e, is mo
dified
; very few ca
n identify ima
g
e
tampers
do
wn to the
gran
ularity of o
n
e
-
pixel le
vel. I
n
some
appli
c
ation
s
,
su
ch
ability could
be
extremely e
s
sential. F
o
r e
x
ample, if an
image i
s
us
e
d
a
s
a
critical
pie
c
e of
evid
ence in th
e
court
or in
a
poli
c
e
investig
ation
,
a ge
neralize
d
an
swer
a
s
t
o
whethe
r th
e entire ima
g
e
or pa
rt of it
is
altered
to
so
me de
gre
e
m
a
y not be
a
c
cepta
b
le by t
he la
w. To
b
e
exa
c
t, it may be m
anda
tory
that the image shoul
d not allow for eve
n
tiny modi
fication ever si
n
c
e it was ta
ken. In this paper,
we propo
se a
n
authenti
c
ati
on schem
e that is able
to detect an
d lo
cate imag
e tampers with th
e
accuracy of one pixel at the be
st ca
se
. This is
don
e by first con
s
tru
c
ting line
a
r blo
c
ks in th
e
image i
n
such a
way that
they interse
c
t with
one
anothe
r in
di
fferent di
re
ctions. Se
co
nd,
a
sign
ature
is created
for ea
ch
blo
c
k for t
he p
u
rp
os
e of
authe
nticatio
n. And
finally,
the
sign
ature
i
s
embed
ded b
a
ck into the image. As a result, ea
ch
pi
xel is prote
c
t
ed by four si
gnatures a
n
d
any
tampered pix
e
l can b
e
pin
pointed by e
x
amining it
s
corre
s
p
ondin
g
sign
ature
s
.
The re
st of the
pape
r is
orga
nize
d a
s
follo
ws. T
he p
r
op
ose
d
tec
hniq
ue is
de
scrib
ed in Se
ction
2, followe
d
by
experim
ental
results in Section 3. Se
ction 4
p
r
e
s
ents a securi
ty analysis. A compa
r
iso
n
of
detectio
n
gra
nularity is sho
w
n in Sectio
n
5. Fi
nally, Se
ction 6 give
s some
con
c
lu
ding re
marks.
2. Proposed
Techniqu
e
2.1. Cons
tru
c
ting th
e
Au
thentic
a
tion Blocks
a
nd Signature Blocks
Without l
o
ss
of gen
erality, we
a
s
sume
that
8-bit gra
y
scal
e
ima
g
e
s
a
r
e
dealt
with. For
other form
ats, the same techni
que ap
pli
e
s, t
oo. The image is first divided into e
qual-si
z
ed
B
B
blocks, whi
c
h
are
refe
rred
to as the
a
u
thentication blocks
. And t
hen, in
ea
ch
authenti
c
atio
n
block, fou
r
se
ts of pixel
s
a
r
e coll
ecte
d
in
four di
re
ction
s
: ho
rizontal,
vertical,
45
and
4
5
wrap-
arou
nd dia
g
o
nals. The
s
e
sets of linear b
l
ocks a
r
e refe
rre
d to as
si
g
nature bl
ocks
. Namely,
H
i
= {
p
ij
|
j
= 0
,
…,
B
1}: horizo
n
tal blo
c
ks,
i
= 0, …,
B
1,
V
j
= {
p
ij
|
i
= 0, …,
B
1}: vertical blo
c
ks,
j
= 0, …,
B
1,
X
m
= {
p
ij
|
i
=
0, …,
B
1,
j
= (
m
+
i
)
m
od B
}: –45
blo
c
ks,
m
= 0, …,
B
1, and
Y
n
= {
p
ij
|
i
= (
B
+
n
j
)
mo
d
B
,
j
= 0, …,
B
1}: 45
block
s
,
n
= 0, …,
B
1,
Whe
r
e
p
ij
de
notes the im
age pixel an
d
mo
d
is the modulo op
eration.
Figure 1 depict
s su
ch a
c
o
ns
tr
uc
tio
n
.
45
w
r
ap
-
a
r
o
u
n
d
d
i
ag
o
n
al
s
i
gna
t
u
r
e
bl
o
c
k
s
A
u
th
e
n
tic
a
tio
n
bl
o
c
k
(
B
B
):
V
e
r
t
i
c
a
l
s
i
gna
t
u
r
e
b
l
ocks
45
w
r
a
p
-
a
r
oun
d
di
a
g
o
n
a
l
s
i
gna
t
u
r
e
bl
oc
k
s
.
. .
.
.
.
Ho
r
i
z
o
n
t
a
l
s
i
gna
t
u
r
e
bl
o
c
k
s
H
0
H
1
V
0
V
1
X
0
Y
0
Figure 1. The
Authenticatio
n Block a
nd
Signature Blocks
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 638
6 –
6392
6388
After establi
s
hing the
sign
ature bl
ocks, a si
gn
ature is cre
a
ted fo
r each
su
ch bl
ock
and
then em
bed
d
ed into
the i
m
age. If the
DES sy
stem i
s
u
s
e
d
for si
g
nature
ge
ne
ration, 64
bits
are
requi
re
d for both the inpu
t and output data. Furthe
rm
ore, if the last two bits
(least-sig
n
ifica
n
t
bits, LSBs) of
each pixel a
r
e use
d
for e
m
beddi
ng,
the size of an authent
ication block
should be
128
128 (i.e.,
B
= 1
28) si
n
c
e the
r
e
are
4
B
sig
nature
blocks. The
r
e
f
ore, the
colle
ction of th
e first
6 bits of each pixel in a signature blo
c
k is ha
sh
e
d
first by su
ch functio
n
s a
s
MD5 o
r
SHA
to
prod
uce a 64
-bit data. And then th
is data is encrypte
d
using the
D
ES system to produ
ce a 64
-bit
sign
ature,
wh
ich i
s
finally e
m
bedd
ed b
a
ck into t
he
LSBs for the
pu
rpo
s
e of a
u
th
enticatio
n. Th
e
above proced
ure is
rep
eate
d
on all sig
n
a
t
ure blo
c
ks.
Whe
n
pe
rforming auth
ent
ication, the
con
s
tr
u
c
tion
of the authe
ntication bl
o
c
ks an
d
sign
ature
blo
c
ks a
r
e
rep
e
a
ted a
s
befo
r
e, followe
d b
y
the same
DES encryptio
n pro
c
e
s
s. Now,
the re
sult
s
can b
e
mat
c
hed
agai
nst
those extr
a
c
ted f
r
om th
e LSBs in t
he ima
ge. T
h
e
mismat
ched blocks will be recorded in
the followi
ng four sets, respectively:
E
H
= {
H
i
|
i
=
i
0
,
i
1
, …,
i
h
1
}: hori
z
ontal
bl
o
c
ks,
E
V
= {
V
j
|
j
=
j
0
,
j
1
, …,
j
v
1
}: ver
t
ic
al
blocks
,
E
X
= {
X
m
|
m
=
m
0
,
m
1
, …,
m
x
1
}: –45
blocks, an
d
E
Y
= {
Y
n
|
n
=
n
0
,
n
1
, …,
n
y
1
}: 45
blo
c
ks,
W
h
er
e
h
,
v
,
x
, an
d
y
a
r
e
the
respe
c
tive numb
e
rs
of
mismat
che
d
blocks. Th
e b
a
si
c id
ea
of the
algorith
m
is that sin
c
e
ea
ch pixel
is
p
r
otecte
d by f
our
sig
nature
s
an
d the
si
gnature bl
ocks
intersect
with
one
an
other,
if a
spe
c
ific
pixel is
ind
e
e
d
tamp
ere
d
with, mi
smatches will
o
c
cur in
all of its four
corre
s
p
ondin
g
sign
atures.
On the
contra
ry, if some of
t
he co
rre
sp
on
ding si
gnatu
r
es
are mat
c
he
d, it can be co
nclu
ded that
the pixe
l has not been alt
e
red. Fig
u
re
2 illustrate
s t
he
algorith
m
of tampe
r
detecti
on in an
auth
enticatio
n blo
ck. Each pixel,
p
ij
(0
≤
i
,
j
≤
B
–1), in a block
is che
cke
d to se
e if it is tampered with. Th
is is
d
one by exa
m
ining its
co
rre
sp
ondi
ng fou
r
sign
ature
s
, i.e., horizo
n
tal, vertical, and
two
diago
nal
ones. If all four si
gnatu
r
e
s
mism
atch, th
e
pixel will be reporte
d as b
e
en tampe
r
ed
with.
for
(
i
= 0, …,
B
–1)
if
(
H
i
E
H
) # Hori
zontal
signatu
r
e mi
smat
che
s
for
(
j
= 0, …,
B
–1)
if
(
V
j
E
V
) # Vertica
l
signatu
r
e mi
smat
che
s
m
←
j
–
i
;
if
(
m
< 0)
m
←
m
+
B
; # Wrap
arou
nd
end if
n
←
i
+
j
;
if
(
n
>
B
–1)
n
←
n
–
B
; # Wrap around
end if
if
(
m
E
X
) and
(
n
E
Y
)
# Both
diago
nal sig
n
a
ture
s mism
a
t
ch
p
ij
: tampered
pixel;
end if
end if
end for
end if
end for
Figure 2. The
Algorithm of Tampe
r
Dete
ction
2.2. Analy
s
is
One of the m
a
in sh
ort
c
omi
ngs of mo
st
other
tamp
er detection te
chni
que
s is t
hat they
usu
a
lly creat
e a sig
natu
r
e
for an im
age
block, an
d if
a mism
atch
o
c
curs, the blo
ck
as
a wh
ole
is
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age Protection by Intersecting Sign
ature
s
(Chu
n-Hung Chen
)
6389
identified a
s
being tamp
ered with. The
r
e is no way o
f
distingui
shin
g whi
c
h pixel (or pixel
s
) i
s
the
victim. The essen
c
e of the
prop
ose
d
scheme lie
s on
the fact that each pixel i
s
p
r
otecte
d by four
intersectin
g
sign
ature
blo
c
ks. When
ever on
e pi
xel
is tampe
r
e
d
with, it causes the f
our
corre
s
p
ondin
g
sig
natu
r
e
s
to mismat
ch
and, thr
oug
h
the interse
c
t
i
ng st
ru
cture,
the tampe
r
e
d
pixel can b
e
easily pin
poin
t
ed. In other words, if
less
than four mi
smatche
s
o
ccur for a pixel,
we
can elimi
nate
the possi
bili
ty of
tamperi
ng. This
met
hod is thu
s
very accurate in identifying
tampered pix
e
ls a
s
well a
s
their lo
cation
s in t
he im
ag
e. There are, however, so
me co
ndition
s in
whi
c
h this
scheme
will make false
positive reports. Fi
gure
3 illustra
tes such
a sit
uation, in whi
c
h
the bla
c
k
pixels
re
pre
s
e
n
t those
pixels that h
a
ve
b
een
altered
b
y
attackers.
The fo
ur sets of
mismat
che
d
blocks
are:
E
H
= {
H
i
1
,
H
i
2
,
H
i
3
},
E
V
= {
V
j
1
,
V
j
2
,
V
j
3
},
E
X
= {
X
m
1
,
X
m
2
,
X
m
3
}, and
E
Y
=
{
Y
n
1
,
Y
n
2
,
Y
n
3
}. It is obvious that, beside
s
the four bla
c
k pixels, the
system will e
r
roneo
usly rep
o
rt
the cente
r
pixel (re
pre
s
e
n
ted by a white
pixel)
as a ta
mpered on
e, i.e., a false positive.
Y
n
1
Y
n
2
Y
n
3
H
i
1
H
i
2
H
i
3
X
m
1
X
m
2
X
m
3
V
j
3
V
j
2
V
j
1
Figure 3. Example of a Fal
s
e Positive
Fi
gure 4.
A
T
a
mpered Regi
on (bla
ck pixels)
Figure 5. Nu
mber of Fal
s
e
Positives vs
Numb
er of Randomly
T
a
m
pere
d
Pixels
Actually, the
numbe
r of ta
mpered pixel
s
(or th
e
size
of the tamp
ered regio
n
)
de
termine
s
the numb
e
r o
f
mismatch si
gnatures. If the form
er in
crea
se
s, the l
a
tter incre
a
ses, too. The
r
e is
no con
s
traint
on the maxi
mum si
ze
of a tampe
r
ed
regio
n
; ho
we
ver, the shap
e of the re
gi
on
doe
s affe
ct the n
u
mbe
r
o
f
false
po
sitives i
n
the
det
ection. Fi
gu
re 4 ill
ustrate
s
a
n
exam
pl
e, in
whi
c
h the bl
ack pixels re
pre
s
ent the tampe
r
ed pi
x
e
ls. As ou
r algorith
m
ide
n
tifies tampe
r
ed
pixels by fo
ur interse
c
ting
sign
ature
s
, th
e set of
repo
rted pixel
s
wi
ll form a
con
v
ex sha
pe
(the
red
polygo
n
i
n
the fig
u
re).
As all
pixels i
n
t
he co
nvex sha
pe are
re
ported
a
s
ta
mpered with, the
unchan
ged p
i
xels (white p
i
xels) are false po
sitive
s. The situatio
n
gets wo
rse if the tampered
pixels spre
ad
rando
mly across the imag
e. The re
su
lt of a simulatio
n
is sh
own Fi
gure 5, in whi
c
h
the tampere
d
pixels are g
enerated ra
n
dom
ly across the authentication blo
c
k (128
128
). It
can
be se
en that
the numb
e
r o
f
false po
sitives in
crea
se
s rapidly a
s
the
numbe
r of tampered pixel
s
increa
se
s, al
most rea
c
hin
g
16,00
0 wh
e
n
the
latter i
s
only a few h
u
ndre
d
s. Su
ch
a phen
ome
n
o
n
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
0
5000
10000
15000
20000
# false
positives
# ta
m
p
ered
p
ixels
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 638
6 –
6392
6390
verifies the a
bove analy
s
is that as the
repo
rted pi
xe
ls form a con
v
ex shape, if the location
s of
the tampe
r
e
d
pixels are randomly
gen
erated, th
e
convex area
will be
co
me
very large, which
results in
a g
r
eat nu
mbe
r
of false p
o
siti
ves.
De
notin
g the num
be
rs of re
po
rted,
tampered, a
n
d
false po
sitive pixels a
s
R
,
T
, and
F
, resp
ectively, the followin
g
rel
a
tionship hold
s
:
R
=
T
+
F
Therefore, when
R
a
c
hiev
es its high
est
(i.e., the si
ze of the bl
ock), in
crea
sing
T
will
certai
nly
d
e
c
r
e
as
es
F
, whi
c
h
expla
i
ns th
e p
eak in Fig
u
re 5.
The
sa
me f
a
lse
po
sitive effect i
s
al
so
expecte
d in
other
existing
blo
c
k-ba
se
d
authe
ntic
atio
n techniq
u
e
s
. In practi
ce,
however, a
s
an
attacker u
s
u
a
lly tries to alter the se
mantics
of the image, tampered pixel
s
tend to cluster
together
(ma
y
be in seve
ral locatio
n
s). Ran
domly alt
e
ring th
e pixe
ls is me
anin
g
l
ess and
hen
ce
not likely to happe
n.
2.3. Handlin
g Irregular Image Sizes
If the image size is n
o
t multiples of the size
of the aut
hentication bl
ock, som
e
thing must
be do
ne fo
r t
he extra
area
s. Since 6
4
b
i
ts are r
equi
red for a
sign
ature, eve
r
y
32 pixel
s
can
be
colle
cted to f
o
rm a
n
indivi
dual a
u
thenti
c
ation
blo
ck. In
those area
s,
ho
we
ver, i
f
the sign
ature
mismat
che
s
,
it can be o
n
ly con
c
lud
ed th
at one or
more pixels in
su
ch a bl
ock
co
uld have b
e
e
n
altered. Fi
gure 6 sho
w
s
su
ch a
situatio
n
.
The granul
a
r
ity of identification i
n
tho
s
e are
a
s is th
us
32 pixels.
p
B
I
ndi
vi
dual
s
i
g
n
at
ur
e
bl
ocks
q
B
Figure 6. Han
d
ling the Situation in whi
c
h
the Image Size is n
o
t Multiples of
B
B
3. Experimental Re
sults
(a)
(b)
(c
)
(d)
Fig. 7: (a) ori
g
inal imag
e, (b) sig
natu
r
e
s
embed
ded (PSNR
= 42.7), (c)
tampe
r
ed image,
(d
)
result of tamper dete
c
tion
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Im
age Protection by Intersecting Sign
ature
s
(Chu
n-Hung Chen
)
6391
The g
r
ayscal
e “Le
n
a
”
ima
ge is u
s
e
d
to
test the prop
ose
d
algo
rith
m: Figure
7(a
)
sh
ows
the o
r
iginal
i
m
age
and
(b
) i
s
the
versi
on
with
sign
a
t
ures em
bed
ded. T
he PS
NR bet
wee
n
the
origin
al an
d t
he em
bedd
e
d
version i
s
42.7, whi
c
h
is quite
a
c
ce
ptable; (c) shows the
im
age
tampered in
Lena’
s rig
h
t e
y
e, and (d) i
s
the result
of tamper d
e
tection. The tam
pere
d
pixels
are
corre
c
tly iden
tified, together with only a few false po
sitives.
4. Securit
y
A
n
aly
s
is
As
th
e D
ES
en
c
r
yp
tion
s
y
s
t
e
m
is
us
e
d
to
gen
erate
the
sig
natures,
the p
r
op
osed
method
is
se
cure a
g
a
i
nst the
attack of m
anip
u
la
ting individu
al
image
pixel
s
. Thre
e othe
r
attacks
are
th
e
sea
r
c
h
,
coll
a
g
e
, and
cut-and-pa
ste
attacks [7, 8], whi
c
h a
r
e co
mmon for bl
ock-wise con
t
ent
authenti
c
atio
n techniq
u
e
s
. Becau
s
e th
e
attacked
ima
ge h
a
s to m
a
i
n
tain g
ood
visual
qu
ality, the
size of the p
a
s
ted bl
ocks h
a
s to b
e
very
small
in
ord
e
r for ke
epin
g
the hom
oge
ne
ity of the block
conte
n
t. Therefore, the
ke
y requireme
n
t
for these
ki
nds of atta
cks to be
su
ccessful is that
the
block si
ze i
s
small en
oug
h
,
usually less
than or eq
ual
to 8
8 pixels. In the propo
sed meth
od,
as
the size of the block is 1
2
8
12
8, it clea
rly make
s the
s
e attacks infeasi
b
le. That
is, even if th
e
attacker may
forge an aut
hentic ima
g
e
from a
data
base co
ntaini
ng hund
re
ds
of thousa
n
d
s
o
f
authenti
c
ima
ges, it will certainly have
poor vi
sual
quality due to block effects and in
co
rrect
block content
. In conclu
sio
n
, our metho
d
is
invulnerable to thes
e attac
k
s
.
5. Comparis
on of De
te
cti
on Granulari
t
y
The dete
c
tio
n
gran
ula
r
ity of the prop
osed
method i
s
compa
r
e
d
with those of Patra
et
al.
’s
[4], Qi
et a
l
.
’s [5], and Wu’
s
[6] m
e
thod
s. Beca
use th
e g
r
an
ularity of Qi
et al.
’s an
d
Wu’
s
method
s de
p
end
s on the i
m
age
size, a unified si
ze
of 256×256 p
i
xels is u
s
ed
in the analysi
s
here. Th
e co
mpari
s
o
n
is shown in Tabl
e
1, in
which it is obvious t
hat our meth
od outpe
rform
s
the others.
Table 1. Co
m
pari
s
on of the
Detectio
n Granula
r
ity
The pro
posed
Patra
et al.
's
Qi
et al.
's
Wu’s
1×1=1 pixel
8×8=64 pixels
8×8=64 pixels
45 pixels
6. Conclusi
on
In this p
ape
r,
we
have
de
scrib
ed
a te
ch
nique to
ide
n
tify tampere
d
pixels in
an
i
m
age. It
is ba
se
d u
p
on dividin
g
t
he ima
ge int
o
authe
nticat
ion blo
c
ks a
nd a
rra
ngin
g
linea
r si
gna
ture
blocks in su
ch a way that they interse
c
t at
every pixel. As
a con
s
e
que
n
c
e, ea
ch pix
e
l is
prote
c
ted by four sign
atures and
such an
a
r
r
ang
em
ent ma
ke
s o
u
r
te
chniq
ue
capabl
e of, in
the
best
ca
se, pi
npointin
g a
si
ngle alte
red
pixel. This te
chni
que
pre
s
erves the p
e
rceptu
a
l si
mila
rity
of the
origi
n
al an
d the
waterma
r
ked
i
m
age
s, a
nd
it is
also
se
cure
ag
ainst
variou
s
po
ssi
ble
attacks. Altho
ugh fal
s
e
po
sitives are li
ke
ly to
be rep
o
rted
if
altered pixels are
spread ran
domly
throug
hout th
e image, an
attacke
r se
ems to have
no rea
s
on t
o
rand
omi
z
e
the alteratio
n
s.
Therefore,
ou
r meth
od i
s
v
e
ry u
s
eful
for protecti
ng
th
e content
s of
the imag
es at
the g
r
a
nula
r
i
t
y
of one pixel.
Referen
ces
[1]
Liu H,
Lin J,
Hua
ng J.
Imag
e auth
entic
atio
n usi
ng co
nten
t based w
a
ter
m
ark
. Proc. IEEE Int. Sy
m.
Circuits a
nd S
ystems. 2005; 4
:
4014-4
0
1
7
.
[2]
Ra
w
a
t
S, Ram
an B. A
ch
aotic
s
y
stem
b
a
se
d
fragi
l
e
w
a
t
e
rm
arkin
g
sch
eme
for ima
ge tam
p
er d
e
tectio
n.
Int. J.
Electron. Commun.
20
11; 65: 84
0-84
7.
[3]
X
i
'an
Z.
A se
m
i
-frag
i
le
d
i
gi
ta
l
wa
te
rm
a
r
ki
ng
a
l
go
ri
thm
i
n
wa
ve
l
e
t tra
n
s
fo
rm
dom
a
in
ba
se
d
on
Arn
o
l
d
transform
.
Pro
c
. IEEE
Int. Conf. Signal Proc
ess. 2008: 2
2
1
7
-22
20.
[4]
Patra JC, Phua JE, Rajan D.
DCT
domain
w
a
termarki
ng s
c
he
me us
ing C
h
in
ese Re
main
der T
heor
e
m
for ima
ge a
u
th
enticati
on.
Pro
c
. IEEE
Int. Conf. Multimedi
a and E
x
p
o
. 201
0: 111-1
16.
[5]
Qi X, Xi
n X,
Chan
g R.
Ima
ge a
u
the
n
ti
cation a
nd ta
mp
er detecti
o
n
usin
g tw
o
compl
e
mentar
y
waterm
arks.
Proc. IEEE Int. Conf. Image Pro
c
ess. 2009: 4
2
57-4
260.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 638
6 –
6392
6392
[6]
Wu
Y.
T
a
mpe
r-Local
i
z
a
t
i
on
W
a
termark
i
n
g
w
i
th Systemati
c
Error C
o
rrec
t
ing C
o
d
e
. Proc. IEEE Int.
Conf. Image Pr
ocess. 200
6: 1
965-
196
8.
[7]
Holl
iman M,
Memon N. C
ounterfe
itin
g Attacks
on Oblivi
ous Bl
ock
-
w
i
se Ind
e
p
e
n
dent Invisi
bl
e
W
a
termarkin
g Schemes.
IEEE Trans. Im
age Process
. 20
0
0
; 9: 432-4
41.
[8]
Barreto PS
LM
, Kim HY, Ri
jmen V. T
o
w
a
rd sec
u
re
p
ublic-k
e
y
b
l
oc
k
w
i
s
e fra
g
il
e
authe
nticati
o
n
w
a
t
e
rmarking.
IEE Proc. Vision, Imag
e an
d Sign
al Process
i
n
g
. 2002; 1
49: 5
7
-62.
[9]
Liu
Q. An A
d
a
p
tive B
lin
d W
a
t
e
rmarkin
g A
l
go
rithm for C
o
l
o
r
Image.
T
E
LKO
M
NIKA Indo
ne
sian
Jour
na
l
of Electrical En
gin
eeri
n
g
. 20
1
3
; 11: 302-
309.
[10]
T
an X, H
u
D.
A W
a
termarkin
g Me
th
od B
a
s
ed
on Optim
i
z
a
tion
Statistics
.
T
E
LKOMNIKA Indo
nesi
a
n
Journ
a
l of Elec
trical Eng
i
ne
eri
n
g
. 201
3; 11: 4
794-
480
2.
Evaluation Warning : The document was created with Spire.PDF for Python.