A Novel Algorithm For Speech Based Authentication Of Digital Images Using DWT.

International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

)" * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,
A %ovel Al,orithm For $peech /ased Authentication 0f
1i,ital Ima,es 2sin, 134.
Atul Dwivedi
1
, Raghvendra Bhushan
2
National Institute Of Technology Raiur, National Institute Of Technology Raiur
atul.!oin"g#ail.co#, raghvendra2$%&"g#ail.co#,

A / $ 4 R A C 4
$ecurity of important data has al-ays 5een an important necessity of human 5ein,# 4raditionally
important used to 5e stored in form of papers# 1evelopment of di,ital computers and computin,
technolo,ies has ena5led data to 5e secured di,itally# 6any al,orithms have 5een developed for
securin, te7t data# 6any secure data transmission applications has 5een developed#
Authentication is used for securin, content for authentication# Authentication can 5e done in
special domain, spectral domain and in hy5rid domain# In this paper -e have proposed a novel
al,orithm to em5ed an audio si,nal into an ima,e usin, discrete -avelet transform and a
fra,mentin, scheme# 4his al,orithm provides a 5etter and faster encodin, and decodin, process
as compared to traditional methods#

Inde7 4erms8 Authentication9 -avelet transform9 speech authentication9 di,it place partitionin,9
:$%R, 6$;

I# I%4R012C4I0%

Digital co##unication and storage technology is i#roving very raidly. It has #ade conversion and
reroduction of intellectual roerty very easy '1(. Authentication is the way to rotect this roerty.
Authentication can )e used to authenticate the owners, license infor#ation, or other infor#ation related
to the digital o)!ect carrying the water#ar*. +ater#ar* can )e used to chec* the irated coy of the data
'2(. I#age Authentication is a very secure Authentication. The Authentication caacity of an i#age is
calculated )y entroy of the i#age.

, - . / log2/ 0 0 0 0 0 0 0 0 0 . 112

+here / is histogra# counts. 3istogra# reresents fre4uency of articular i5el in an i#age. ,ntroy
can )e used to chec* which i#age 16over i#age2 is #ost suita)le for hiding i#ortant infor#ation. It is
erfor#ed on two i#age shown. In ca#era#an i#age entroy is 7.%% and in circuit i#age entroy is
8.$&, therefore first i#age is )est suita)le for i#age Authentication.

In Authentication, a controlled distortion introduced into an i#age with a roer algorith#. This
algorith# is called encoding algorith# whereas for authentication of the i#age a decoding algorith# is
used. I#age is large data)ase of i5els. In this data)ase, encoding algorith#s usually identify redundant
data. This redundant data is relaced with the water#ar* data. 9everal algorith#s have )een roosed in
this field in satial do#ain, fre4uency do#ain and in hy)rid do#ain. In satial do#ain encrytion is on
i5el values, in sectral do#ain in transfor# coefficients and hy)rid do#ain, encrytion are )ased on
certain sche#es. In this aer, we shall first discuss different traditional #ethods for digital i#age
Authentication then roosed sche#e of authentication. At last we have resented resulted o)tained
after i#le#entation of roosed sche#e.
International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

)) * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,

II# $:A4IA< 106AI% A24=;%4ICA4I0%

In satial do#ain digital i#age Authentication is done )y #odifying the i5el values of the host i#age
':(. It was started with #odifying least significant )it lane of an i#age. To level authentication is done
)y #odifying ;9B lane and 2nd lane near to the ;9B lane. This is done )y considering the fact that
;9B lane has very less significance in the i#age. A R<B color i#age consist of three grayscale i#ages
and an A greyscale i#age 1generally stored as unsigned = )it integer2 is collection of = )inary i#ages.
Bit lanes of a greyscale i#age can )e searated into )it lane. This concet can )e #ore efficient if two
near)y )it lanes are #odified in encrytion. The #a!or disadvantage of ;9B techni4ue is its oor
i##unity to #aniulations. >ig1 reresents ;9B coding techni4ue. It can )e o)served fro# the i#age of
;9B and 2
nd
lane that they do not contain i#ortant infor#ation. I#ortant infor#ation is stored in
7
th
and =
th
lane which are near to ?9B. It can )e concluded that ;9B and )it lanes near to ;9B can )e
used for encoding. >ig 11i2 reresents all the )it lanes co#)ined. Then fig 1 1*2 reresents all )it
lanes co#)ined with ;9B lane relaced )y fig1 1!2.



1a2 1)2 1c2 1d2


1e2 1f2 1g2 1h2

1i2 1!2 1*2

Fi,ure 18 $eparation of 5it planes for <$/ Codin, (a) <$/ 5it plane (5)
nd
5it plane (c) &
rd
5it plane (d)
"
th
5it plane (e) )
th
5it plane (f) >
th
5it plane (,) 7
th
5it plane (h) '
th
5it plane (6$/) (i) All 5it planes
com5ined (.) -atermar?(?) 3atermar?ed ima,e

III# $:;C4RA< 106AI% A24=;%4ICA4I0%

?any ti#es Authentication is also done in sectral do#ain, li*e in discrete >ourier transfor# 1D>T2
do#ain '&(, co#lete i#age discrete cosine transfor# 1D6T2 do#ain '@( and )loc*Awise D6T do#ain '8(.
In D6T do#ain Authentication the coefficients are #odified as follows

International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

)> * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,
xiB-xiC#a5'xiom ,sign1xi2Ðik( 00000000. 122

+here,
xi is selected coefficient and om is noise sensitivity inde5,
Di is the 4uantiDation ste and the value of * is in )etween @ and 8. By this Authentication #a5i#u#
acco##odated /9NR for rando# noise is 22.1 dB..
In literature, it is said that in transfor# do#ain it is )etter than secial do#ain. These are found to )etter
in ter#s of visi)ility and security of the water#ar*. In satial do#ain, D6T has )een )etter than the D>T
in ter#s of security. 9o#e author has clai#ed that Authentication should )e done at low fre4uencies '2(
'&(, so#e of the# say that Authentication should )e done at higher fre4uencies '@(, '7(. In fact in so#e of
the literatures it has )een roved that ro)ust water#ar* can )e o)tained using adatively so#e sectral
co#onents in so#e evolutionary #anner.
IV# =@/RI1 A24=;%4ICA4I0%

In transfor#ing to the fre4uency do#ain, ti#e infor#ation is lost. +hen loo*ing at a >ourier transfor# of
a signal, it is i#ossi)le to tell when a articular event too* lace. Ti#ing infor#ation is very i#ortant
in various i#ortant signal li*e )eginning of the signal, end of the signal, a)rut changes etc. In an effort
to correct this ro)le#, Dannis <a)or 11$&82 had ta*en >ourier transfor# to analyDe a very s#all
fraction of the signal at a ti#e, this techni4ue is called windowing techni4ue. This transfor# is called
9T>T. The 9T>T reresents a sort of co#ro#ise )etween the ti#eA and fre4uencyA)ased views of a
signal.

+avelet analysis reresents the ne5t logical steE a windowing techni4ue with varia)leAsiDed regions.
+avelet analysis allows the use of long ti#e intervals where we want #ore recise low fre4uency
infor#ation, and shorter regions where we want high fre4uency infor#ation '=(. >or #any signals, the
lowAfre4uency content is the #ost i#ortant art. It is what gives the signal its identity. The highA
fre4uency content, on the other hand, is not very i#ortant. 6onsider a hu#an voice If you re#ove the
highAfre4uency co#onents, the voice sounds different, )ut you can still tell whatFs )eing said. 3owever,
if you re#ove enough of the lowAfre4uency co#onents, you hear gi))erish. In our authentication syste#
we have ta*en these advantages into account for e#)edding seech into facial i#ages using digit lace
artitioning.>ig. 2 reresents difference )etween satial do#ain aroaches and sectral do#ain
aroaches, It can )e easily seen that wavelet analysis also have scaling infor#ation that is i#ortant for
identifying i#ortant art of a signal or i#age.







Fi,ure 8 1ifference 5et-een different spectral and hy5rid analysis techniAues
International Journal of
)7 * + !1", IJAFRC All Ri,hts Reserved
+avelet transfor# has )een roved to )e #ore efficient in ter#s of security and ro)ustness of
authentication syste#s.
V# :R0:0$;1 $C=;6;

In general a seudo rando# se4uences are used as a *ey for e#)edding
i#rove#ent a recogniDed signature can )e water#ar*ed invisi)ly. But still )ecause these are not
encoded they are vulnera)le to attac*s. 3ere we roose an algorith# to encode seech signal of a
ersonFs na#e as encoding *ey which can rovide #ore security and ro)ustness in the security of
authentication syste#.


















Fi,ure&8 :rocess of Authentication of ima,es usin, speech si,nal
>ig. : reresents a flowchart of authentication using wavelet analysis. 9eech of
i#ages are to )e authenticated are stored in a data)ase then loaded to generate a te#late using novel
algorith# resented in the co#ing section. Then : level D+T is erfor#ed over the te#late generated.
In the arallel rocess an i#age of sa#e erson is ta*en and rocessed using D+T and then e#)edding
of water#ar* is done using e4uations

Authentication involves e4uation 1:2. ;et us say i#age I to )e authenticated using water#ar* + then
authenticated i#age I+ is given )y



3ere Ii and +i can )e low fre4uency 1;;2 or low high 1;32 fre4uency co#onent and *G %.
VI# $:;;C= 4;6:<A4; B;%;RA4I0%
Recorded digital seech signal are store in for# of an
for#ation of an i#age, here we ta*e first @@12@ ele#ents of sound in an array and then convert it into a
#atri5 of 2&@H22@ele#ents. The ele#ents of array are ta*en and their digit laces are searated.
the #agnitude is a five digit nu#)er and so searating the nu#)er into its digit lace gives five nu#)ers.
International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"#

+ !1", IJAFRC All Ri,hts Reserved
+avelet transfor# has )een roved to )e #ore efficient in ter#s of security and ro)ustness of
In general a seudo rando# se4uences are used as a *ey for e#)edding water#ar* into an i#age, as an
i#rove#ent a recogniDed signature can )e water#ar*ed invisi)ly. But still )ecause these are not
encoded they are vulnera)le to attac*s. 3ere we roose an algorith# to encode seech signal of a
which can rovide #ore security and ro)ustness in the security of
Fi,ure&8 :rocess of Authentication of ima,es usin, speech si,nal
>ig. : reresents a flowchart of authentication using wavelet analysis. 9eech of
i#ages are to )e authenticated are stored in a data)ase then loaded to generate a te#late using novel
algorith# resented in the co#ing section. Then : level D+T is erfor#ed over the te#late generated.
of sa#e erson is ta*en and rocessed using D+T and then e#)edding
of water#ar* is done using e4uations
Authentication involves e4uation 1:2. ;et us say i#age I to )e authenticated using water#ar* + then
authenticated i#age I+ is given )y,
- C 000000000000000001:2
3ere Ii and +i can )e low fre4uency 1;;2 or low high 1;32 fre4uency co#onent and *G %.
$:;;C= 4;6:<A4; B;%;RA4I0%
Recorded digital seech signal are store in for# of an array. It has to )e converted into a #atri5 for the
for#ation of an i#age, here we ta*e first @@12@ ele#ents of sound in an array and then convert it into a
22@ele#ents. The ele#ents of array are ta*en and their digit laces are searated.
the #agnitude is a five digit nu#)er and so searating the nu#)er into its digit lace gives five nu#)ers.
Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&
---#i.afrc#or,
+avelet transfor# has )een roved to )e #ore efficient in ter#s of security and ro)ustness of
water#ar* into an i#age, as an
i#rove#ent a recogniDed signature can )e water#ar*ed invisi)ly. But still )ecause these are not
encoded they are vulnera)le to attac*s. 3ere we roose an algorith# to encode seech signal of a
which can rovide #ore security and ro)ustness in the security of
Fi,ure&8 :rocess of Authentication of ima,es usin, speech si,nal
>ig. : reresents a flowchart of authentication using wavelet analysis. 9eech of the ersons whose
i#ages are to )e authenticated are stored in a data)ase then loaded to generate a te#late using novel
algorith# resented in the co#ing section. Then : level D+T is erfor#ed over the te#late generated.
of sa#e erson is ta*en and rocessed using D+T and then e#)edding
Authentication involves e4uation 1:2. ;et us say i#age I to )e authenticated using water#ar* + then
000000000000000001:2
3ere Ii and +i can )e low fre4uency 1;;2 or low high 1;32 fre4uency co#onent and *G %.
array. It has to )e converted into a #atri5 for the
for#ation of an i#age, here we ta*e first @@12@ ele#ents of sound in an array and then convert it into a
22@ele#ents. The ele#ents of array are ta*en and their digit laces are searated. 3ence
the #agnitude is a five digit nu#)er and so searating the nu#)er into its digit lace gives five nu#)ers.
International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

)' * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,
Two nu#)ers of the a#ong the searated digits of array ele#ent are used to for# a nu#)er which is
allotted to one of the three #atrices 1say B2 corresondingly the other two nu#)er 1searated fro# the
array ele#ents digits2 are used to for# another nu#)er which is allotted to the corresonding ele#ent
of the other #atri5 1say <2. Now only one digit is left, )ut we need one #ore digit to indicate whether the
sound ele#ent is ositive or negative for that either 1 or Dero can )e used to for# tenth digit of a nu#)er
with the left out nu#)er 1@
th
digit of array ele#ent2 at ones lace and this is allotted to a #atri5 1say B2.
3ence : ele#ents of three #atrices 1R<B2 are o)tained corresondingly ne5t nu#)er of sound array can
)e used to for# other ele#ents of 1R<B2 #atrices corresondingly. If the data is a four digit nu#)er then
Deroes can )e aended in R<B #atri5 values to for# three two digit nu#)ers. 3ere the IRF #atri5 also
shows weather the ele#ent of sound array is ositive or negative. A si#le rogra# can )e #ade in
which R #atri5 can reresent the ositive or negative ele#ents of sound array, if the ele#ent is negative
then tens lace of corresonding R #atri5 ele#ent will )e 1 else %. The ones lace of r #atri5 ele#ent
will )e engendered )y )ra*ing the sound array ele#ent into digit laces and fro# the ten thousand lace
of the corresonding sound array ele#ent the ones lace will )e for#ed. < and B constitute of nu#)er
for#ed )y thousands, hundreds and tens, ones lace of the ele#ent of sound array to for# 2 two digit
nu#)ers. But in conversion of i#age to sound data tyes #ust )e *et in #ind, as an i#age is dou)le,
uint= etc. data tyes which is different fro# that of a sound array 1int182. 9o the data should )e
converted first into suita)le for# )efore the conversion starts.

;et sound array ele#ent )e 18@:2. >or trans#ission, ele#ent are first )ro*en into digit laces )y ta*ing
)y dividing the# )y the 1%%%% for 1%%%%
th
lace 1%%% for 1%%%
th
lace and so. Their corresonding #od
is ta*en and then the lower value of integer is selected and stored in a :J2 #atri5 say 6,

CC D! 19 > )9 & E

Then RC 1!FC11GC1, BC1!FC1GC, /C1!FC&1GC&

The first ele#ent of 6 is Dero reresenting the ositive nu#)er. 3ad )een the ele#ent A18@:2, 6 would
)e

CC D1 19 > )9 & E
The sound array ele#ent )e 12:&.
CC D! !9 1 9 & "E

Then, RC 1!FC11GC1, BC1!FC1GC, /C1!FC&1GC&

>or decoding inverse oeration is done. The digit laces of R, <, B are again searated and stored in a D
#atri5 of order:J2

>or e5a#leE RC! 1, BC)>, /C&" (i#e# data)

Then in decoding sche#e D- '% 1K @ 8K : &(
And the array ele#ent would )e
If 111C !, then
@ (I) CG (1!!!! H 11 G 1!!! H 11 G 1!! H 1 G 1!H1&1G1H1&)
If 111C1, then
@ (I) C ( (1!!!! H 11 G 1!!! H 11 G 1!! H 1 G 1!H1&1G1H1&)

International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

)I * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,



Fi,ure "8 Flo- 1ia,ram for ;ncryption Fi,ure )8 Flo- 1ia,ram for 1ecryption
>igure & deicts the co#lete rocess of encrytion. Any ele#ent L1i2 is )ro*en down to its digit laces
an then those digits are used to for# three nu#)ers for red, green and )lue fra#es which are #erged
together to for# a R<B i#age.
>igure @ deicts the rocess of decrytion of sound fro# the encryted i#age. In this the i#age is again
searated into its R, <, B fra#es and the corresonding ele#ents of fra#e are #erged together and signs
are attached to the# fro# d #atri5 to for# a single nu#)er which is the corresonding nu#)er of sound
array, After that #atri5 is converted into array.
VII# R;$2<4$ A%1 A%A<@$I$
The algorith# e5ecuted in ?AT;AB 7.1%.% software, in a core i@ 4uad core rocessor and windows =
oerating syste#. 9ound was recorded and i#age te#late was generated according to a)ove stated
algorith#. This te#late was encoded into an i#age using D+T and was decoded successfully.
International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

>! * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,

Fi,ure > 8 0ri,inal $ound Fi,ure 78 Ima,e 4emplate Benerated From $ound


Fi,ure '8 Ima,e /efore ;ncryption Fi,ure I8 Ima,e After ;ncryption

Fi,ure 1!8 1ecoded 4emplate Fi,ure 118 $ound 1ecoded From 4emplate
A)ove figures, show the all the i#ages generated at various stes of rocess fro# te#late generation,
encrytion to decrytion and o)taining the original sound.
>or analyDing this sche#e we have ta*en facial i#ages of 1% different ersons as shown in ta)le1 and
alied this algorith# to these i#ages. 9eech signals of every erson has )een recorded as shown in
ta)le1.
$#%0 0ri,inal ima,es Correspondin, voices 4emplates ;ncrypted ima,es
1




International Journal of
>1 * + !1", IJAFRC All Ri,hts Reserved



&

"

)

>

7

'

I

International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"#

+ !1", IJAFRC All Ri,hts Reserved
















Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&
---#i.afrc#or,








International Journal of
> * + !1", IJAFRC All Ri,hts Reserved
1!

4a5le 18 Represents ori,inal ima,es and correspondin, voices of people, the ,enerated templet and the
ima,e in -hich the template -as encrypted usin,
Authenticated i#ages with te#lates have
i#age has )een generated as shown in a)ove ta)le1
The e5eri#ents were carried out on a)ove ten i#ages for hiding in an audio signal. /erfor#ance
evaluation factors li*e /9NR and ?9, values were calculated )y adding various noises 1such asE salt M
eer, <aussian, oison, sec*le2 to the encryted i#age with original i#ages of corresonding erson
as reference and grahs were lotted.

Fi,ure 18 :snr Analysis 0f 1! Ima,es 4a?en For ;7perimentation

International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"#

+ !1", IJAFRC All Ri,hts Reserved


Represents ori,inal ima,es and correspondin, voices of people, the ,enerated templet and the
ima,e in -hich the template -as encrypted usin, 134
Authenticated i#ages with te#lates have )een done using wavelet transfor# and
n generated as shown in a)ove ta)le1.
re carried out on a)ove ten i#ages for hiding in an audio signal. /erfor#ance
evaluation factors li*e /9NR and ?9, values were calculated )y adding various noises 1such asE salt M
eer, <aussian, oison, sec*le2 to the encryted i#age with original i#ages of corresonding erson
as reference and grahs were lotted.
Fi,ure 18 :snr Analysis 0f 1! Ima,es 4a?en For ;7perimentation
Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&
---#i.afrc#or,

Represents ori,inal ima,es and correspondin, voices of people, the ,enerated templet and the
)een done using wavelet transfor# and an authenticated
re carried out on a)ove ten i#ages for hiding in an audio signal. /erfor#ance
evaluation factors li*e /9NR and ?9, values were calculated )y adding various noises 1such asE salt M
eer, <aussian, oison, sec*le2 to the encryted i#age with original i#ages of corresonding erson

Fi,ure 18 :snr Analysis 0f 1! Ima,es 4a?en For ;7perimentation
International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

>& * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,

Fi,ure 1&8 6ean $Auare ;rror Analysis 0f 1! Ima,es 4a?en For ;7perimentation


VIII# C0%C<2$I0%

<rah for ?9, analysis of water#ar*ed i#age and noisy i#age which indicates that the ?9, of salt and
eer noise has #ini#u# value it give conclusion that it affects the water#ar*ed i#age hardly. The
/9NR of salt and eer noisy i#age is high. /9NR is the ea* signal to noise ratio, in deci)els, )etween
two i#ages. This ratio is used as a 4uality #easure#ent )etween two i#ages. If /9NR ratio is high then
i#ages are )etter of 4uality. It shows that the effect of this noise is less on water#ar*ed i#age.

IJ# R;F;R;%C;$

'1( ?in +u, Bede ;iu ,N ?ulti#edia data 3idingN 9ringer, I9BNE $7=A1A&&1$A2$$&A@, . 11$A1:8,
2%%:, International Oournal of Recent Trends in ,ngineering, Pol. 1, No. 2, ?ay 2%%$.
'2( 3artung >, Qutter ?, R ?ulti#edia water#ar*ing Techni4ue,S I,,, roceeding on 9ignal
/rocessing,S vol.=7, No.7, .1%7$A11%7, Ouly 1$$$
':( R. <nana!eyara#an, Q./rasadh, Dr.Ra#ar, RAudio encrytion using higher di#ensional chaotic
#aS International Oournal of Recent Trends in ,ngineering, Pol. 1, No. 2, ?ay 2%%$.
'&( O. O. Q.T O Ruanaidh, +. O. Dowling, and >. ?. Boland, R/hase water#ar*ing of digital i#ages,S
in/roc. Int. 6onf. I#age /rocessing 1I6I/2, vol. :, 9et. 1$$8, . 2:$U2&2.
'@( ?. Barni, >. Bartolini, P. 6aellini, A. /iva, and >. Rigacci, RA ?.A./. identification criterion for
D6TA)ased water#ar*ing,S in /roc. ,uro. 9ignal /rocessing 6onf. 1,V9I/6O F$=2, Rhodes, <reece,
9et. 1$$=.

International Journal of Advance Foundation and Research in Computer (IJAFRC)
Volume 1, Issue 7, July !1"# I$$% &"' ( "')&

>" * + !1", IJAFRC All Ri,hts Reserved ---#i.afrc#or,
'8( R.<nana!eyara#an, Q./rasadh, Dr.Ra#ar, RAudio encrytion using higher di#ensional chaotic
#aS International Oournal of Recent Trends in ,ngineering, Pol. 1, No. 2, ?ay 2%%$..
'7( >. Deguillau#e, <. 6sur*a, O.TO Ruanaidh, and T. /un, RRo)ust :D D>T video water#ar*ing,S in
I9MTW9/I,Fs 11th Annu. 9y#., ,lectronic I#aging F$$E 9ecurity and +at
'=( ?. Barni, >. Bartolini, P. 6aellini, A. /iva, and >. Rigacci, RA ?.A./. identification criterion for
D6TA)ased water#ar*ing,S in /roc. ,uro. 9ignal /rocessing 6onf. 1,V9I/6O F$=2, Rhodes, <reece,
9et. 1$$=.
'$( A!ay.B.<adichal 12%112. Audio +ave 9teganograhy. International Oournal of 9oft 6o#uting and
,ngineering 1IO96,2 11@2E .17&A178.
'1%( R. 9afa)a*hsh, 9. Xa)oli, A. Ta)i)iaDar, RDigital +ater#ar*ing on 9till I#ages Vsing +avelet
Transfor#,S 12%%&2 /roc. I,,, International 6onference on Infor#ation TechnologyE 6oding and
6o#uting IT66F%&, vol.1, . 871A87@.