TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol.12, No.7, July 201
4, pp
. 5399 ~ 54
0
7
DOI: 10.115
9
1
/telkomni
ka.
v
12i7.537
3
5399
Re
cei
v
ed
De
cem
ber 1
5
, 2013; Re
vi
sed
Jan
uar
y 21, 2
014; Accepte
d
February 1
8
, 2014
Resear
ch of Blind Forensics Algorithm on Digital Image
Tampering
Jin Hong
y
i
n
g
Comp
uter coll
e
ge of Chi
na W
e
st Normal U
n
i
v
ersit
y
Shid
a Roa
d
No
. 1, ShunQing
District,
Nanch
ong, Sich
ua
n, 637
00
2, Chin
a
e-mail: ji
nh
ong
yi
ng
_c
w
n
u@1
63.com
A
b
st
r
a
ct
W
i
th rapid
de
velo
p
m
ent of
the inter
net a
nd
the
mu
lti
m
edi
a techn
o
lo
gy, the dig
i
tal
image
s
tamp
eri
ng w
i
th blind fore
nsic
s technol
ogy b
e
co
me a n
e
w
researc
h
direct
ion in the stud
y of informati
o
n
se
cu
ri
ty. Th
e te
ch
no
l
o
g
y
i
s
b
a
s
ed
o
n
th
e
d
i
g
i
ta
l
im
a
g
e
,
a
n
d
i
t
i
s
on
l
y
to
rea
l
i
z
e
ima
g
e
i
n
te
g
r
i
t
y
and
authe
nticity of t
he certif
ic
atio
n, so it is w
i
d
e
ly
used
in th
e civ
i
l fiel
ds suc
h
as
new
s rep
o
rts, judici
al
pro
o
f an
d
mi
litary fie
l
ds s
u
ch as
military
intel
lig
enc
e an
alysis. So
th
e r
e
searc
h
of it h
a
s gre
a
t sig
n
ifi
c
ance
an
d bro
a
d
app
licati
on pr
o
s
pect. In the
pap
er,
the exp
l
orati
on an
d re
search o
n
the
paste tamp
eri
ng an
d splic
in
g
tamp
eri
ng w
i
t
h
bl
in
d fore
ns
ics al
gorith
m
are
ma
de.
T
h
e bl
in
d fore
ns
ics al
gorith
m
base
d
o
n
ra
di
al
Kraw
tchouk co
py-an
d
-past
e
i
n
vari
ant
mo
me
nt is pro
pos
ed.
As the curr
ent
copy-a
nd-
past
e
bl
ind
forens
i
cs
alg
o
rith
m has
low
locali
z
a
ti
on accur
a
cy, and p
oor
rob
u
stness of the post-proc
es
sing in so
lvin
g th
e
prob
le
m. Bas
e
d o
n
sl
id
ing
w
i
ndow
b
l
ock
ma
tching
meth
od
and
the
rad
i
a
l
Kraw
tchouk i
n
varia
n
t
mo
me
n
t, it
prop
oses a c
o
py-an
d
-past
e
tamperi
ng w
i
th
blin
d fore
ns
ics
alg
o
rith
m. T
h
e exp
e
ri
me
ntal
results sh
ow
that
the a
l
gor
ith
m
c
an
effectively
l
o
cate th
e ta
mpere
d
ar
ea,
an
d it
has v
e
ry s
t
rong r
obust
n
e
ss of the
rotati
n
g
oper
ation, JPE
G
compress
io
n
,
Gaussian n
o
i
s
e, etc.
Ke
y
w
ords
: bli
nd forens
ics al
gorith
m
, di
gital
imag
e, tamp
eri
ng, researc
h
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
With the
ra
pi
d develo
p
me
nt of multime
d
ia
an
d di
gital technolo
g
y, digital ph
otos
have
become the
indispen
sabl
e
nece
s
sities
of life. Tr
aditional film pictures
are i
n
stead by digit
a
l
photo
s
, whi
c
h show that
the “digital
image
er
a” ha
s a
rrived
.
In ord
e
r t
o
a
c
hieve t
he
enha
ncement
purpo
se of t
he digital
im
age visual
effect, a la
rge
numbe
r of im
age p
r
o
c
e
s
si
ng
and editin
g
software,
su
ch
as Photo
s
ho
p, A
CDSee,
li
bra
r
ie
s
have been widely use
d
.
Ho
wev
e
r,
advan
ced technolo
g
y brin
gs the co
nve
n
ien
c
e in dai
ly
life, at the
same time, they also bri
ght
some
hidd
en
trouble
in
mode
rn life.
With t
he
wid
e
ly appli
c
atio
n of the ima
ge p
r
ocessin
g
softwa
r
e, the
tamperin
g o
f
digital image be
come
s more conve
n
ient, its effect is al
so very
reali
s
tic,
so
some
p
eople
with
mali
cio
u
s
motives sprea
d
fo
rgin
g digital
ima
ges in
order to
achi
eve ulte
ri
or p
u
rp
ose.
Whe
n
tamp
ered
with di
gita
l image
is
wi
dely appli
ed t
o
medi
a repo
rts,
sci
entific
re
search
and th
e co
urt, it will
no d
oubt
aff
e
ct the
norm
a
l order
of society an
d
ca
use
damag
e to pe
rso
nal ri
ghts [
1
-3].
2. Digital Image Fore
n
si
cs
The digital im
age fore
nsi
cs technolo
g
y can be divided
into active and passive fo
ren
s
ics
techn
o
logy.
Active fore
nsics technolo
g
y is e
m
be
d
ded in
the
media i
n
a
d
v
ance
indi
ca
tive
informatio
n, and ma
ke a
u
t
henticatio
n o
f
embedd
ed
i
n
formatio
n. The existing
a
c
tive foren
s
i
c
s
techn
o
logy m
a
inly inclu
d
e
s
digital wate
rmarki
ng
and
digital si
gnatu
r
e, etc. Th
ese
techni
que
s a
r
e
adopte
d
by the ba
sic tho
ught of emb
eddin
g
or
ad
d addition
al informatio
n a
u
thenticity a
nd
integrity of digital image id
entification.
2.1. The Digital Signatur
e
The digital
si
gnature tech
nology is
also kn
own
as
electroni
c si
g
nature, it is
attache
d
according
to
the ca
ble
s
in
elect
r
oni
c fo
rm, and
the
conte
n
t is
used to id
entify the si
gnatu
r
e
identity data.
Whe
n
use the digital sign
ature to i
denti
f
y the authenticity of digital
media, it sho
u
ld
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5399 – 54
07
5400
extract the
ori
g
inal me
dia d
a
ta inform
atio
n in
adva
n
ce
and
kee
p
it in
storage, thro
ugh
storin
g th
e
informatio
n to identify the object informat
i
on to make t
he identification of the obje
c
t.
2.2. Digital Wa
termar
king.
Digital wat
e
rmarking a
c
co
rding to its tol
e
ran
c
e to im
age chan
ges,
it can be divided into
the fragile
watermark
[4-7], half a fragile
waterm
a
r
k
and
r
o
bu
s
t
wa
ter
m
a
r
k
[8]. Fragile watermark
and fragil
e
waterma
r
k a
r
e
both apply th
e authenti
c
ity and
integrity of accurate in
the certificati
on,
the differe
nce of them i
s
that
when
contai
ning watermark medium
i
s
cha
nge
d,
fra
g
ile
watermarkin
g
are
ea
sily da
maged, th
en i
t
is ha
rd
to b
e
dete
c
ted; A
nd semi-f
ragil
e
wate
rma
r
king
can withsta
n
d
rea
s
on
able certain
extent
of distortio
n
, and the u
n
re
aso
nabl
e dist
ortion
will ca
u
s
e
further d
e
teri
orate.
Robustness
of watermark is ha
rd to remove and
has a
stro
ng anti-interference
ability,
and it ca
n al
so abl
e to wi
thstand
all so
rts of commo
nly editing proce
s
s and
waterma
r
k
attack
tools. It is mai
n
ly use
d
for
copyri
ght info
rmation ide
n
tification
of di
git
a
l wo
rks. As f
o
r e
a
ch
type
of
digital wate
rmarking, thei
r basi
c
fram
ewo
r
k i
s
mai
n
ly include
s
three pa
rts:
embed
ded p
a
rt,
transmissio
n cha
nnel, and
extracting p
a
rt. Embedd
ed termin
al is embe
dde
d
with the digital
media
kn
own
as pa
rt of
the id
entificati
on info
rm
atio
n; Tra
n
smissi
on
cha
nnel
i
s
the
carrie
r
of
identificatio
n
informatio
n, it can
pa
ss pa
ram
e
ters an
d
key i
n
formatio
n to extra
c
ting
part;
Extraction p
a
r
t ca
n extra
c
t informatio
n to
identify t
he o
b
ject, an
d the
n
acco
rdin
g to the extra
c
ti
ng
accuracy
and
co
mplete
ness of
the i
n
formation m
a
ke
the ide
n
tificat
i
on a
nd
auth
enticatio
n of t
h
e
objec
ts
.
3. O
v
er
v
i
e
w
of Blind Digi
tal Image Fo
rensics Te
ch
nolog
y
3.1. Definitio
n
and Clas
sification
o
f
Bl
ind Digital Image Foren
sics
Blind digital image foren
s
i
cs
refers to the pro
c
e
s
s o
f
using emb
e
dded info
rma
t
ion to
identify authenticity of ima
ge and
obtai
n evidence; the whol
e pr
o
c
e
ss is in
dep
ende
nt of any
sign
ature
or
premi
s
e i
n
formation [9]. Despite th
e cu
rre
nt digital i
m
age m
anip
u
l
ation technol
ogy
has b
een very mature, the fake effect o
f
tamperi
ng o
f
digital image is not ea
sy to cause visual
differen
c
e
s
a
m
ong
peo
ple,
but tam
peri
n
g op
eratio
n
will
inevitably cause cha
nge
in
the statisti
ca
l
feature
s
of digital image [1
0].
Image bli
nd
foren
s
ics i
s
t
o
use the
st
atis
tical
ch
aracteri
stics
ch
ange
s of im
age a
n
d
make
the
det
ermin
a
tion
of the a
u
thenti
c
ity, integr
ity, and
primitivism of the
ima
ge [11
-
12]. Bl
ind
foren
s
ics te
chnolo
g
y ha
s
no
spe
c
ial
re
quire
ment
s
f
o
r im
age
a
c
q
u
isition
device, com
p
a
r
ed
with
the active forensi
c
s tech
no
logy, it also do not
need to
add any ima
ge authe
ntica
t
ion informati
on,
so it ha
s im
portant p
r
a
c
tical value. Bli
nd fo
re
nsi
cs technol
ogy i
s
mainly for image
cont
ent
tamperi
ng an
d for the evidence coll
ectio
n
,
No
w, the image blind forensi
c
s techn
o
logy can b
e
divided into the followi
ng three
c
a
te
go
r
i
es
:
1. In view of the image
con
t
ent authenticity di
scrimin
ation: the main purp
o
se is to judge
wheth
e
r initial
l
y acquired i
m
age
s are su
ffered with
so
me form of proce
s
sing o
r
tampe
r
ing.
2. In view of
the imag
e source id
entificat
ion: it can
judge th
e i
m
age d
a
ta a
c
qui
sition
device, thi
s
ki
nd of technol
ogy will
con
n
e
ct ima
ges with the co
mm
on featu
r
es i
m
age
sou
r
ce,
in
orde
r to match the image
s to a parti
cul
a
r type of the source devi
c
e.
3. Analyses t
he hidd
en im
age fore
nsi
cs: As for the integrity of image tampe
r
in
g is an
importa
nt informatio
n se
curity techn
o
lo
gy bran
ch
, so use it a
s
a
part of the i
m
age foren
s
i
cs.
Comp
ared wi
th hidden an
alysis te
chnol
ogy, image hi
dden de
epe
r analysi
s
fore
nsi
cs n
eed the
se
cret info
rm
ation extracti
on.
As for
digital
image tam
p
e
r
ing o
p
e
r
atio
ns, the
com
m
on tamp
eri
ng me
ans are thro
ugh
the copyin
g o
f
original ima
ge and m
a
ski
ng sp
ecif
i
c
ta
rget area of i
m
age, thu
s
create the
sce
n
e
whi
c
h a
r
e n
o
t
in the origi
nal imag
e scene o
r
hid
e
some i
m
po
rtant target, th
e purpo
se of
this
tamperi
ng
m
ean
s i
s
the
copy-an
d
-pa
s
te ima
ge tam
perin
g [13
-
1
4
]. The tam
p
e
r
ing
metho
d
has
many obviou
s
adva
n
tage
s, beca
u
se th
e co
py-an
d
-p
aste p
a
rt of i
m
age h
a
s
no
obviou
s
visu
al
differen
c
e in
terms
of brig
htness, col
o
r,
noise
with the ori
g
inal i
m
age, therefore, the tamp
ere
d
image loo
k
s realisti
c, it's hard to judge f
r
om t
he visio
n
. And due to its operation is simple; the
curre
n
t copy-and-pa
ste ta
mperi
ng
ha
s
a very
wid
e
r
ange
of a
ppli
c
ation
s
. In thi
s
a
r
ticl
e, we
will
put forwa
r
d
a
ne
w
algo
rith
m to im
prove
the
blin
d
forensi
c
s d
e
tect
ion effe
ct a
n
d
robu
stne
ss of
the algorith
m
[15-17].
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of Blind Fore
nsi
cs Algo
rithm
on Digital Im
age Tam
perin
g (Jin
Hon
g
yi
ng)
5401
4. The Model
of Cop
y
-and
-pas
te Tamp
ering
Copy-and
-pa
s
te p
e
rfo
r
m
tamperi
ng
o
peratio
n i
n
t
he
sam
e
im
age, the
o
p
e
ration
i
s
usu
a
lly adopt
ed co
pying m
e
thod. The
p
a
ste an
d cop
y
part have n
o
interse
c
tion
or overl
ap in
the
image an
d po
sition deviatio
n
is also obvi
ous. From
th
e operation p
r
ocess of cop
y
-and
-pa
s
te, the
temperi
ng im
age shoul
d h
a
ve the following feature
s
:
(1) T
he same
image ha
s the same o
r
si
milar conn
ect
ed are
a
;
(2) Di
spl
a
ce
ment of the identical o
r
sim
ilar
are
a
is often gre
a
ter th
an a certai
n thre
shol
d
value;
(3)
The a
r
e
a
of these
are
a
s
is
often gre
a
ter t
han
a ce
rtain threshol
d, whi
c
h i
s
th
e are
a
is
large
eno
ugh
. Durin
g
the
study, res
earchers often to
simplify the re
se
ar
ch q
u
e
s
t
i
on,
if
we ma
ke
the hypothe
sis that the t
a
mpe
r
ed im
a
ge is
only o
ne area h
a
s been
copi
e
d
and
paste
d to
anothe
r area
in the image.
4.1. Simple
Cop
y
-Mobile-Paste Re
gio
n
Tampering
Model
Simply co
py-mobile
-pa
s
te
regi
on ta
mp
ering
refers t
o
the
are
a
to
be
co
pied
o
n
ly after
displ
a
cement
of sim
p
le o
peratio
n, an
d
paste
it dire
ctly to othe
r
area
in the
i
m
age, n
o
t af
te
r
pro
c
e
ssi
ng o
peratio
ns
su
ch as scali
ng,
rotation,
the
r
efore, copy
area a
nd
st
i
c
ky
st
ic
k ar
ea
is
equal in the a
r
ea, just exist
s
po
sition de
viati
on betwe
en co
py area
and pa
ste area. For this type
of Tampe
r
ing
,
its model is
copy-mobil
e
-paste ta
mp
eri
ng model, ju
st as sho
w
n in
the Figure 1.
x
y
1
()
F
I
2
()
F
I
()
F
I
Figure 1. Cop
y
-move-p
a
ste
Regio
n
Tam
per Mo
del
As sh
own in
the Figure 1, the origin
file is
(,
)
F
xy
, after tamperin
g the file ca
n b
e
expre
s
sed a
s
'
(,
)
F
xy
, and the followin
g
type ca
n be obtain
e
d
which is expressed a
s
(1
):
2
'
2
(,
)
(
,
)
(,
)
(,
)
(
,
)
F
xy
xy
I
Fx
y
F
xx
y
y
x
y
I
(1)
Whe
r
e the
x
and
y
represe
n
t the displ
a
ceme
nt of the
x
and
y
, and the
1
I
rep
r
e
s
ent
s the copi
ed are
a
,
2
I
repre
s
e
n
ts the pa
sted
area a
nd th
e two area i
s
the sa
me,
(,
)
F
xy
repre
s
e
n
ts the gray valu
e of position
(,
)
x
y
of the file an
d
'
(,
)
F
xy
repre
s
e
n
ts the gray
value of positi
on
(,
)
x
y
after tampering.
4.2. Cop
y
-Paste Area Ta
mpering Mo
del throug
h Specific Pro
cessing O
p
e
r
ations
Thro
ugh
ce
rtain processin
g
of
co
py-an
d
-pa
s
te a
r
e
a
tamper
with t
he mod
e
l, also kn
own
as
copy-t
ran
s
form
-mo
b
ile-paste ta
mpe
r
ing m
odel,
it refers the
copi
ed
regi
on after
po
st-
pro
c
e
ssi
ng o
peratio
n su
ch
as scaling, rotation,
and then move it paste
d into other a
r
ea
s of the
image
of the t
a
mpe
r
with th
e op
eratio
n.
Acco
rdi
ng to
the tampe
r
in
g
pro
c
e
s
s, Th
e
tampe
r
ing
can
not only cau
s
es the po
sitio
n
offset between co
py
are
a
and the paste area, at the same time, the
two area
will meet certain
transfo
rmation relationship between r
egi
ons. The tam
pering model
is
as sho
w
n in
Figure 2 and
Figure 3.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5399 – 54
07
5402
1
()
F
I
2
()
F
I
x
y
1
()
F
I
2
()
F
I
x
y
Figure 2. Cop
y
-rotate
-pa
s
te Regi
on Ta
mper
Model after
Certain Op
eration
Figure 3. Cop
y
-zoo
m-p
a
ste
Regio
n
Tam
per
Model after
Certain Op
eration
Acco
rdi
ng the model, the
origin pictu
r
e
(,
)
F
xy
and tempe
r
ed pi
cture
'
(,
)
F
xy
can be
expre
s
sed a
s
(2):
2
'
2
(,
)
(
,
)
(,
)
(
(
,
)
(,
)
(
,
)
(,
)
F
xy
xy
I
Fx
y
TF
x
x
y
y
h
x
y
n
x
y
x
y
I
(2)
Whe
r
e,
(,
)
F
xy
denotes origi
n
pict
ure an
d
'
(,
)
F
xy
den
otes tempe
r
e
d
picture, he
1
I
rep
r
e
s
ent
s the copi
ed area
,
2
I
rep
r
e
s
ent
s the pa
sted are
a
. In
the model, when the
picture is
zoom
ed, the area i
s
not the same, an
d the
T
rep
r
e
s
ent
s the tran
sform of the zoo
m
and
rotation.
(,
)
hx
y
and
(,
)
nx
y
rep
r
e
s
ent
s the other op
era
t
ions.
5. Radial Kra
w
tc
h
ouk Co
p
y
and Paste Blind Fore
n
sics Algori
t
hm of In
v
a
riant Momen
t
s
5.1. Algorith
m
Theor
y
Aiming at a
kind of comm
on imag
e forensi
c
s
copy-and-pa
ste ta
mper
i
ng, it propo
se
s a
blind foren
s
ics algo
rithm b
a
se
d on radi
al Krawt
c
h
o
u
k
mome
nt invariant
s. The
main ide
a
of the
detectio
n
alg
o
rithm is lo
ca
l matching. I
n
the firs
t s
t
ep, it us
es
the wavelet trans
f
orm to extrac
t
the low freq
uen
cy comp
onent of ima
ge, and then
it the extract invariant ra
dial Krawt
c
h
ouk
moment of low frequ
en
cy comp
one
nt
and the feature vector of co
m
positio
n ch
ara
c
teri
stic m
a
trix
are
dictio
nary
so
rted. It ca
n re
alize the
matchin
g
of chara
c
te
risti
c
s
pie
c
es
toget
her co
uple
d
with
simila
r threshold, blo
c
ks
of sp
acin
g a
nd the
ar
e
a
threshold, it
finally debu
g
s
math
emati
c
al
morp
holo
g
y to determi
ne the final
co
py-and-pa
ste tamperi
ng area
.
Thre
e Points
of Algorithm are a
s
belo
w
:
(1)Throug
h
DWT p
r
o
c
e
s
sing
with th
e
su
sp
ect im
age, a
nd ext
r
actin
g
ima
g
e
of lo
w
freque
ncy
su
bban
d, it can
greatly redu
ce the
num
b
e
r of ima
ge b
l
ock, and the
low fre
que
ncy is
not sen
s
itive to noise at the same time, whi
c
h
can en
han
ce the extracte
d features ro
bu
stne
ss.
(2) The
mai
n
idea
s
of cop
y
-and
-pa
s
te t
a
mpe
r
d
e
tecti
on i
s
lo
cal
m
a
tchin
g
, that i
s
m
a
ke
wavelet
su
bb
and
blo
c
k in
low fre
quen
cy of the
im
age, a
nd th
e
n
ma
ke
the
image fe
ature
extraction, an
d then match
all of the extracte
d
image
features, wh
en the matchi
ng error is le
ss
than a ce
rtain
thresh
old, it can b
e
co
nsi
dere
d
a su
ccess match, it is the co
re of the algo
rithm.
(3) As for th
e sele
ction o
f
image cont
ent feature
s
. The radial
Krawt
c
ho
uk
moment
invariant feat
ure of im
age
bloc
k, copy-and-pa
ste ta
mper
with
th
e dete
c
ting i
m
age
s, be
ca
use
Krawt
c
ho
uk
moment
s is b
a
se
d on
cla
s
sical discr
ete
Krawt
c
ho
uk p
o
lynomial. Ra
dial Kra
w
tch
o
u
k
can
well d
e
s
cribe
the
i
m
age
feature inva
riant
moment, a
n
d ha
s the
good
di
sting
u
ish
identificatio
n
feature
s
. S
e
co
nd, the
radial Kr
awt
c
hou
k inva
ria
n
t moment
also
ha
s rotation
invarian
ce, th
e tampe
r
i
s
throu
gh
with t
he copy-and
-paste
metho
d
, and it
can
rep
r
o
c
e
s
s th
e
operation. Fo
r insta
n
ce as one of the m
o
st comm
on
copy-rotatio
n-displ
a
cement
-pa
s
te mod
e
, it
not only ca
n
enhan
ce th
e
visual effect
of modificati
on, and h
a
rd
er to dete
c
t, becau
se
som
e
algorith
m
can
not resi
st rotation o
peration p
e
rfo
r
man
c
e. Th
erefore
,
the ra
dial K
r
awt
c
ho
uk h
a
s
good d
e
script
i
on of image
rotation invari
ant moment f
eature
s
fo
r im
age tamp
er d
e
tection. Ba
sed
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of Blind Fore
nsi
cs Algo
rithm
on Digital Im
age Tam
perin
g (Jin
Hon
g
yi
ng)
5403
on the above
analysi
s
, in the pap
er, it select
s t
he ra
dial Kra
w
tch
o
u
k mom
ent i
n
variant feat
ure
image blo
c
k, and imag
e co
py-and
-p
aste
tampering
m
e
thod with bli
nd foren
s
i
cs
algorith
m
. Flow
cha
r
t of the algorithm i
s
as
sho
w
n in the
Figure 4.
DW
T
Figure 4. Flow Ch
art of the Algorithm
5.3. Application of Discr
e
t
e Wav
e
let Transf
orm in Redu
cing th
e Amoun
t of
Image Data
Dete
ction
ba
sed
on
blo
c
k
matchin
g
alg
o
rithm i
n
solv
ing the
key p
r
oble
m
s,
mai
n
ly have
two point
s: o
ne is to redu
ce the time
complexity;
the other i
s
ho
w to extract t
he feature. In th
e
pape
r, testing
image with di
screte wavele
t Transfo
rm
[15] (DWT, Di
screte Wavel
e
t Tran
sform
)
is
prop
osed . Wavelet tran
sform i
s
a kind of time domain, fre
q
uen
cy domai
n or airsp
a
ce -
freque
ncy
do
main tran
sfo
r
mation;
hav
e at the
sa
me time
do
main a
n
d
freque
ncy
do
main
locali
zation
p
r
ope
rtie
s. In the algo
rithm
of wa
velet,
the feature
e
x
traction p
e
rf
orme
d in lo
w
freque
ncy
pa
rt, it can
gre
a
t
ly redu
ce the
numb
e
r
of image
blo
ck, l
o
w frequ
en
cy is n
o
t se
nsiti
v
e
to noise at th
e same time,
whi
c
h can en
han
ce ro
bu
stness of the e
x
tracted featu
r
es.
In the rece
nt years, wavel
e
t analysi
s
h
a
s
be
en dev
elope
d very rapidly, its ap
plicatio
n
area
s i
n
cl
udi
ng ima
ge
proce
s
sing,
co
mputer re
co
g
n
ition, si
gnal
pro
c
e
s
sing,
and m
any ot
her
fields, discret
e wavelet de
compo
s
ition (DWT) can de
comp
ose the
two-di
men
s
io
nal image
sig
nal
into a lo
w fre
quen
cy ap
pro
x
imation su
b
band
and l
e
ve
l of detail, vertical
and
di
agon
al detail
s
an
d
three hig
h
fre
quen
cy su
bb
and. The Lo
w frequen
cy ap
proximatio
n subba
nd mea
n
s the maximu
m
scale of
the
optimal app
roximation of the
o
r
iginal
i
m
age
thro
ug
h
wavelet de
comp
ositio
n unde
r
the minimum
resolution, its statistical
ch
ara
c
teri
stic
s i
s
simila
r to original imag
e, and mo
st of the
energy is fo
cued on th
e subba
nd ima
g
e
; While the
high fre
que
n
c
y su
bban
d i
s
mainly ima
g
e
detailed info
rmation in different resol
u
tion and diffe
rent scales, th
e lowe
r the resol
u
tion is, there
will be the
hi
gher
co
ntain
s
informatio
n. Images
after
wavelet multi
-re
sol
u
tion d
e
com
p
o
s
ition
in
the low frequ
ency pa
rt will still keep the
overview
of the sou
r
ce im
age and
spa
c
e features, the
loss of hig
h
frequ
en
cy det
ail inform
atio
n ca
n be
ne
glecte
d. The
image afte
r
discrete
wav
e
let
decompo
sitio
n
results a
s
shown in figure 3-4.
And di
screte wavel
e
t transfo
rm can be
rega
rded
as u
s
in
g lo
w frequ
ency fi
lter and
high
frequ
ency fi
lter to de
co
mpose the i
m
age into
lo
w
freque
ncy
an
d high
freq
u
ency
coeffici
ent, as
fo
r t
w
o-dime
nsio
nal si
gnal, d
i
screte
wavel
e
t
decompo
sitio
n
and re
co
nst
r
uctio
n
proce
ss
c
an b
e
expre
s
sed a
s
shown in Figu
re 5.
Figure 5. Disc
rete
Wavelet Trans
f
orm
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5399 – 54
07
5404
In the pap
er,
it propo
se
s
copy-an
d
-pa
s
te tampe
r
ing
with blin
d forensi
c
s alg
o
rit
h
m, firstly
it throug
h di
screte
wavelet
tran
sform (DWT) a
s
p
r
e
-
p
r
ocessin
g
fo
r the figu
re,
a
nd extra
c
t th
e
discrete
wav
e
let wra
p
the low-f
r
eq
ue
ncy su
bba
nd
as optimal
approximatio
n of the image,
throug
h the a
bove the scattered wavelet
transfo
rm
an
alysis, it can
be found the
that the size
o
f
the lo
w fre
q
u
ency
com
pon
ent of o
r
igin
al
image
can
b
e
re
du
ced
to
a qu
arte
r of j
( j i
s
fo
r wave
let
decompo
sitio
n
serie
s
), through
the p
r
o
c
e
ssi
ng
it
ca
n a
c
hieve th
e
goal
of
woul
d of redu
cin
g
the
amount
of im
age d
a
ta. Th
en, ba
sed
on
the wavelet coefficient
s, lin
e slid
e bl
ock
and the
blo
ck to
can
be
extracted
to the
ra
dial K
r
awtchou
k
win
d
o
w, and then in turn it
c
a
n turned into
sub
s
e
que
nt match a
c
tion.
6. Results of Experiment
and An
aly
s
is
In the exp
e
ri
ment, in o
r
de
r to valid
ate t
he effe
ctiven
ess of th
e alg
o
rithm, in
the
first st
ep,
we
sele
ct the
gray l
e
vel of
256 i
n
the
e
x
perime
n
t an
d the
size of
image
s i
s
5
12×512
as th
e
exper
imental images
, if
t
he image is
R
G
B type
of
image, it can be
c
onver
ted to grayscale
image
s. In the sel
e
ctio
n of image
th
re
sh
old, acco
rdin
g to the expe
rimental meth
od, the sele
ction
of similarity thre
shol
d is
set as R
= 0.0
03; the
si
ze o
f
the image b
l
ock is 8
×
8, t
he sp
aci
ng bl
ock
threshold i
s
L
= 12; Image area th
re
shol
d is S >
512
×512
×0.85%. In the ex
perim
ents, we
sele
ct
pictures tan
k
as te
st ima
g
e
s
, an
d te
st thi
s
al
gorit
h
m
re
spe
c
tively in
copy
- di
spl
a
ceme
nt - pa
st
e
,
addin
g
n
o
ise, co
py-rota
t
ion-
di
spla
ce
ment-p
aste,
mirrori
ng operation
s
and JPEG
lossy
comp
re
ssion
ca
se
s, and th
e accuracy of
experime
n
ta
l
testing, the erro
r rate a
nd
the analysi
s
o
f
misjud
gme
n
t rate are cal
c
u
l
ated.
6.1. Cop
y
- d
i
splacemen
t - pas
t
e Tam
p
er De
tec
t
io
n
(a) O
r
igin pi
ct
ure
(b) Pict
ure aft
e
r tampe
r
ing
(c) Prelimin
ary detection i
m
age
(d)
Re
sults af
ter morpholo
g
ical p
r
o
c
e
ssi
ng
Figure 6. Orig
in Dete
ction
Re
sults
As can b
e
se
en through
th
e re
sult
s of e
x
perime
n
t, the algo
rithm a
bout joini
ng t
a
mpe
r
ing
with ima
g
e
s
after Ga
ussia
n
noi
se
dete
c
tion
effect i
s
goo
d a
s
sho
w
n in
the
Fig
u
re
7, with
th
e
increa
se of n
o
ise inte
nsity, the area
of the test match
gradu
ally narrowed.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of Blind Fore
nsi
cs Algo
rithm
on Digital Im
age Tam
perin
g (Jin
Hon
g
yi
ng)
5405
(a) Pict
ure aft
e
r tampe
r
ing
(b) Dete
ction image
Figure 7. Multi-co
py-p
aste
Dete
ction
6.2. Tamper Detec
t
ion of
Adding Nois
e
In orde
r to test and verify the robu
stne
ss of al
gorithm, different p
o
st-proce
ssin
g
operation
s
a
r
e pe
rform
ed
according
to
the tampe
r
in
g imag
e, su
ch as ad
ding
different level
s
of
Gau
ssi
an , Noise
(SNR
=
45 dB, 35
dB, 25 dB, 15
d
B
),
the test
re
sults
as
sh
own as i
n
figu
re
8 ,
r
e
spec
tively.
As ca
n be se
en throu
gh th
e experim
ent, the detectio
n
effect of algorithm a
bou
t joining
tamperi
ng im
age
s after a
dding G
a
u
s
sian noi
se
is good, but with the increasi
ng of no
ise
intensity, the area of the te
st
match i
s
gradually na
rro
wed.
(a) SN
R
=
45
d
B
(b) SN
R
=
35
d
B
(c
) SNR
=
2
5
d
B
(d) SN
R
=
15
dB
Figure 8. Det
e
ction
Re
sult
s
of Adding G
aussia
n
Noi
s
e
6.3. Ev
aluation Index Cal
c
ulation o
f
the Algorith
m
In the p
ape
r,
copy-and
-p
aste tampe
r
ing
with
the
alg
o
rithm of blin
d f
o
ren
s
i
c
s eval
uation
index has b
e
en introd
uce
d
; in the spe
c
ific calcul
ation the error
rate and mi
sj
udgme
n
t rate
o
f
each pi
cture,
on the
first st
ep, it
rando
m
l
y sele
ct from
a
re
ctangle
i
n
the
co
py, a
nd p
a
ste
it into
same
ima
ge i
n
the
re
gion
of disj
oint, th
e si
ze
s
of co
py are
a
a
r
e
the 3
2
×32, 6
4
×
64,
96
×96
a
nd
128
×12
8
. We
tamper with
the image
s with som
e
p
o
st-p
ro
ce
ssin
g operation
s
, such as a
d
d
i
ng
Gau
ssi
an noi
se, JPEG co
mpre
ssion, th
e statistics
of its accuracy u
nder
sp
e
c
ific post-processi
ng
operation
s
an
d misjud
gme
n
t rate.
1) Pro
c
e
ssi
ng
accura
cy an
d misjud
gme
n
t rate, error rate
In the
repli
c
a
t
ion area
und
er th
e
conditi
on of
proce
s
sing
op
eratio
ns, p
a
ste
the
imag
e
dire
ctly to another a
r
ea, the detec
tio
n
effect of the tampered imag
e of erro
r rate
and false a
r
e
as
sho
w
n
in
Tab
l
e 1,
and
thro
ugh th
e Fi
gure 9,
the
i
n
tui
t
ive re
sults wi
th the
ch
ang
e of th
e tam
p
er
with the are
a
, the chan
ge o
f
detection re
sults a
r
e
sho
w
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 7, July 201
4: 5399 – 54
07
5406
Table 1. Accura
cy, Inaccu
racy an
d Misd
iagn
osi
s
Rat
e
of Copy-p
a
s
te Tamp
er
Tampering a
e
ra
Accuracy
Error
rate
Misjudgment ra
te
32×32 0.9644
0.0677
0
64×64 0.9723
0.0519
0
96×96 0.9794
0.0405
0
128×128
0.9903
0.0172
0
2) Accu
ra
cy
and mi
sjud
g
m
ent rate,
error rate after j
o
ining
differe
nt intensity of
Gau
ssi
an
noise
Join
in th
e i
m
age
with
G
aussia
n
n
o
ise of 1
5
dB,
25 dB, d
b
3
5
, 45
dB, re
spe
c
tively,
usin
g the
re
spectively
Cha
p
ter al
go
rithm in th
e
det
ecting
imag
e
tampe
r
ing
with accu
ra
cy
and
error rates. Fi
gure 9, a
)
is a
c
cura
cy rate
,
b) is
the error rate, c
)
is
fals
e rate.
(a) G
a
u
ssi
an
noise (dB) aft
e
r addi
ng different
Gau
ssi
an noi
se dete
c
tion
accuracy
(b) G
a
u
ssi
an
noise (dB) aft
e
r addi
ng different
Gau
ssi
an noi
se dete
c
tion
error rate
(c) Gau
s
sian
noise (dB) aft
e
r addi
ng different
G
aussi
an noi
se dete
c
tion misj
udg
ment rate
Figure 9. Accura
cy, Inaccu
racy an
d Misdiagn
osi
s
Rat
e
of Adding G
aussia
n
Noi
s
e with Differe
nt
Intensities
In the chapte
r
, the result
s of copy-a
nd-paste
tamp
eri
ng algo
rithm detectio
n
effect have
been q
uantita
t
ive analyzed
. The experi
m
ents a
r
e m
a
inly aim for copy-displa
c
e
m
ent-p
aste,
and
add G
a
u
ssi
a
n
noi
se, JPE
G
co
mpressi
on. Spin Tu
rn
, mirro
r an
d o
t
her tamp
er
with the type
test,
as
ca
n
be
se
en from th
e e
x
perime
n
tal d
e
tection
ima
g
e
s. In
the
pa
per, it th
e
ob
vious te
st
re
sults
of put forward the alg
o
rith
m ha
s be
en
achi
eved. Th
en, the qu
anti
t
ative analysi
s
is
obtain
ed,
this
cha
p
ter m
a
in
ly calculate
s
error
rate
s
and mi
sju
d
g
m
ent rate
of
detectio
n
a
c
cura
cy in
e
a
ch
tampered
alg
o
rithm, fu
rthe
r results sho
w
the
e
ffe
ctivene
ss of th
e
algo
rithm
an
d the
alg
o
rith
m
has go
od
performan
ce. A
c
cording
to th
e expe
rime
nt, the alg
o
rith
m for
Gau
s
si
an n
o
ise, JP
EG
comp
re
ssion
and othe
r po
st-p
ro
ce
ssi
ng
to tamper
with the image
of the opera
t
ion have very
stron
g
rob
u
st
ness, esp
e
ci
ally for rotating post
-
processing tamp
ering, it
has
stron
g
dete
c
tion
ac
cur
a
cy
.
6. Conclusio
n
In the study, it firstly introd
uce
s
the
cop
y
-and
-pa
s
te t
a
mpe
r
with t
he mod
e
l, co
mbined
with the m
o
d
e
l, a ra
dial Krawtchou
k
co
py-and
-p
as
te
tamperi
ng
wi
th blind fo
ren
s
ics al
go
rith
m of
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Re
sea
r
ch of Blind Fore
nsi
cs Algo
rithm
on Digital Im
age Tam
perin
g (Jin
Hon
g
yi
ng)
5407
invariant mo
ments i
s
pro
posed.
Experiments sho
w
that
the
prop
os
e
d
algo
rith
m can effe
ctively
locate the ta
mperi
ng of copy-an
d
-pa
s
te area,
an
d solve the pro
b
lem of the simila
r ope
ra
ting
algorith
m
whi
c
h i
s
un
able
for resi
st
rot
a
tion, an
d it
also
ha
s very stro
ng
ro
b
u
stne
ss
with
the
JPEG lossy compressio
n a
nd Gau
s
sian
noise and
so
on.
Referen
ces
[1]
Z
hu BB, S
w
a
n
s
on MD, T
e
w
f
i
k
AH. W
hen seei
ng is
n'
t beli
e
vin
g
.
IEEE SignalProcess
ing Maga
z
i
ne,
200
4; 21(2): 40
-49.
[2]
W
u
Qong. F
a
cing the truth
detectio
n
bl
ind
digita
l ima
ge
forensics meth
od rese
arch.
Univers
i
t
y
o
f
defens
e techn
o
lo
g
y
20
08.
[3]
NT
Chang SF
, Lin CY, et al.
Passive-b
lin
d
Image F
o
re
nsi
cs.
Multimedi
a
Security T
e
ch
nol
ogi
es for
Digita
l
Ri
ghts,
W
Z
eng, Yu H, Lin C Y(e
d
s.), Elsvier. 20
06.
[4]
Lan
h T
V
, Cho
ng KS, Emm
a
nue
l S,
et al.
A Survey
o
n
Digita
l
Ca
mer
a
Imag
e F
o
r
e
nsic M
e
thods
.
Procee
din
g
s of
IEEE Internationa
l Conf
erenc
e on
Multim
edi
a and E
x
p
o
, Beiji
ng, Ch
ina. 2
007: 16-
19.
[5]
Z
hou
Lin
na, M
r
W
ang. D
i
gita
l ima
ge for
ens
ics tech
n
o
l
o
g
y
,
Beij
in
g: Bei
jin
g un
iversit
y
of posts a
n
d
telecomm
unic
a
tions press. 20
08.
[6]
Yao Z
u
o
l
i
ang,
W
u
Qiong,
Li
Guohu
i, T
u
Dan.
A recov
e
rab
l
e a
nd
active
d
i
gital
sig
natur
e
appr
oac
h fo
r
imag
e: RADS.
In Proceedings
of IEEE
conference. Beijing, Ch
ina. 2004:
541-546.
[7]
Z
hong
Hu
a, j
i
ao
li c
hen
g.
A di
gital
sig
n
a
t
ure sch
eme f
o
r ima
g
e
cont
ent ver
i
ficatio
n
.
Journ
a
l of
computers,
20
03: 26(6): 7
08-
708.
[8]
Yang Y
i
xia
n
Ni
u Xin
Xi
n. Dig
ital
w
a
term
arkin
g
theor
y a
nd te
chno
log
y
. Be
iji
ng: hi
gher
edu
cation
pres
s
.
200
6.
[9]
Re
y C, Du
gel
a
y
J. A surve
y
o
f
W
a
termark
ing
Algorithms for
Image Auth
enti
c
ation. Eur
a
sip
Journ
a
l o
n
Appl
ied Si
gn
al
Processi
ng,
20
02; 6:61
3-62
1.
[10]
Yeun
g MM, Mintzerf.
An Invisib
l
e W
a
ter
m
ark
i
ng T
e
c
h
niq
ue for Ima
ge Verific
a
tion
.
International
Confer
ence
on
Image proces
sing. Santa B
a
rbara, CA, US
A. 1997; (2): 680-6
83.
[11] Fridrich
J.
Se
curity of F
r
agi
le Auth
entic
ati
on W
a
ter
m
ark
s
w
i
th Local
i
z
ation.
4t
h Co
nferenc
e o
n
Securit
y
an
d W
a
termarki
ng of Multimed
ia C
o
ntents. San Jo
se, CA, USA. 2002: 69
1-7
00.
[12]
F
r
idrich J. Methords for D
e
te
cting Ch
an
ges
in Dig
ital Ima
ges.
Multi
m
e
d
i
a
and S
e
curity
W
o
rkshop a
t
ACM Multim
edia.
Orlando, F
l
o
r
ida, USA. 199
9: 29-33.
[13]
Lin
C, C
h
a
ng
S. Semi-F
ra
gile Watermarking fo
r A
u
the
n
tic
a
ting
JPEG Vi
sual
Co
ntent.
Security and
W
a
termark
i
n
g
of Multimed
ia
Conte
n
s
. San
Jose, CA, USA. 2000: 40-
51.
[14]
Van Sch
y
n
d
e
l
RG,
T
i
rkel AZ
,
Osborne C
F
. A digital w
a
termark.
Proc. Int.Conference in Image
process
i
ng. 1
9
94; (2): 86-90.
[15]
Che
ng E
n
, Z
h
ang
Ro
ng-
Xi
n, Yua
n
F
e
i. A
u
toma
tic D
e
te
ction a
n
d
Ass
e
ssment S
y
st
em of W
a
ter
T
u
rbidit
y
b
a
se
d on Imag
e Processi
ng.
T
E
L
K
OMNIKA Indones
ian J
ourn
a
l of
Electrica
l
Engin
eeri
n
g
.
201
3; 11(3): 15
06-1
513.
[16]
Xi
on
g Ji
e. D
i
gi
tal Me
dica
l Im
age
En
hanc
ed
b
y
w
a
ve
let Il
l
u
min
a
tion-
Refl
ection
Mod
e
l.
TEL
K
OMNIKA
Indon
esi
an Jou
r
nal of Electric
al Eng
i
ne
eri
ng.
2013; 1
1
(1): 1
9
-27.
[17]
Ye T
i
an, Chan
ge Z
h
e
ng, Qiu
hon
g Ke. A Bu
bbl
e
Detecti
on
Algorit
hm Base
d on Sp
arse a
n
d
Red
u
n
d
a
n
t
Image Proces
sing.
T
E
LKOMNIKA Indon
esi
an Jour
na
l
of Electrical E
ngi
neer
ing.
2
013;
11(6): 29
83
-
299
0.
Evaluation Warning : The document was created with Spire.PDF for Python.