Innovative Multi-Level Secure Steganographic Scheme based on Pixel Value Difference

In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
DOI:10.5121/ij cs .2012.2601 1
Inno a i e Mul i-Le el Secu e S eganog aphic
Scheme based on Pixel Value Di e ence
Ma ghny H. Mohamed
1
, Naziha M. Al-Aid oos
2
, and Mohamed A. Bama a
3
1
Depa men o Compu e Science, Facul y o Compu e s and In o ma ion,
Assiu Uni e si y, Egyp .
[email p o ec ed]
2
Depa men o Compu e Enginee ing, Facul y o Enginee ing and Pe oleum,
Hadh amou Uni e si y, Yemen.
[email p o ec ed]
3
Depa men o Compu e Science, Facul y o Science, Hadh amou Uni e si y,
Yemen.
[email p o ec ed]
A
BSTRACT
S eganog aphy is one o he b anches o in o ma ion secu i y ield, i aims o hide in o ma ion in
un ema kable co e media so as no o a ouse an ea esd oppe ’s suspicion. The sec e message is hidden
in such a way ha no signi ican deg ada ion can be de ec ed in he quali y o he o iginal image. The aim
o his pape is o in oduce an e icien s eganog aphic scheme o hide da a o e g ay scale images. This
scheme is based on he p ope y o he human eye, which is mo e sensi i e o he change in he smoo h a ea
han he edge a ea using pixel alue di e ence, besides employing he LSB subs i u ion echnique as a
undamen al s age. The expe imen al esul s show ha he p oposed me hod could success ully achie e he
goals o he high embedding capaci y and main aining he isual quali y, in addi ion, p o ides mo e secu e
da a hiding using selec i e pixel posi ions de e mined by a sec e image (i.e. key). Mo eo e , based on ha ,
he sec e message is eplaced wi h dynamic LSBs, ou scheme can e ec i ely esis se e al image
s eganalysis echniques.
K
EYWORDS
Da a Secu i y, S eganog aphy, Da a Hiding, Leas -Signi ican Bi (LSB), Pixel Value Di e ence (PVD).
1.
I
NTRODUCTION
The de elopmen in echnology and ne wo king has posed se ious h ea s o ob ain secu ed da a
communica ion. In o de o keep he unau ho ized use away, a ie y o echniques ha e been
p oposed. In o ma ion hiding and c yp og aphy a e wo main ways o secu ed communica ion.
S eganog aphy is one o he b anches o in o ma ion hiding, i is he a o hiding in o ma ion by
conceal i s exis ence, unlike c yp og aphy echnique which abou p o ec ing he con en o
message by ans o m i o be meaningless and unin elligible o unau ho ized use s.
An in o ma ion hiding sys em is cha ac e ized by ha ing h ee di e en aspec s ha con end wi h
each o he as shown in Figu e 1: capaci y, secu i y, and obus ness. Capaci y e e s o he amoun
o da a bi s ha can be hidden in he co e medium, secu i y ela es o he abili y o an
ea esd oppe o de ec he hidden in o ma ion easily, and obus ness is conce ned abou he
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
2
amoun o modi ica ion he s ego medium can esis be o e an ad e sa y can modi y o des oy
he hidden in o ma ion [21].
Figu e 1. In o ma ion-hiding sys em ea u es.
S eganog aphy is he a and science o communica ing in such a way which hides he exis ence
o he message, as de ined by Cachin [5]. I includes a a ie y numbe o me hods o hiding da a
in media. These media we e in isible inks, mic odo s, e c. Nowadays, s eganog aphy uses ex ,
images, audio, and ideo media.
Image s eganog aphy is he mos widely used, compa ed wi h he o he ypes o s eganog aphy,
his popula i y because he la ge amoun o edundan in o ma ion p esen in he images ha can
be easily al e ed o hide sec e messages inside hem, and because i can ake ad an age o he
limi ed powe o he human isual sys em (HVS) [2].
In Image s eganog aphy, he e m co e image is used o desc ibe he image used o hold sec e
in o ma ion inside, while he esul ing image which ob ained by embedding sec e da a in o co e
image is called he s ego image.
An image in a compu e is an a ay o numbe s ha ep esen ligh in ensi ies a a ious pixels,
hese pixels make up he image’s as e da a. Digi al images a e s o ed in ei he 24-bi ( ue colo
images) o 8-bi pe pixel iles. Hence 8-bi colo images, like GIF iles, can be used o hide
in o ma ion. He e, each pixel is ep esen ed as a single by e, and he pixel's alue is be ween 0
and 255. G ey scale images a e p e e ed because he shades a e changed e y g adually be ween
pale e en ies, his inc eases he image's abili y o hide in o ma ion [13].
A lo o ways a e used o hide in o ma ion inside he images. Leas signi ican bi (LSB)
subs i u ion, and masking & il e ing echniques a e he mos well known echniques using in
image s eganog aphy. LSB echnique is a simple app oach o hide in o ma ion in an image, bu
any image manipula ion can des oy he hidden in o ma ion in his image easily.
The s eganog aphy domain is g owing up e y quickly, se e al o p ac ical ials and
ma hema ical pape s a e published cons an ly, some o hese pape s a e used LSB-based me hod
in an a emp o ge a new da a hiding echniques o o de elop an exis ing echniques, such as:
[4], [6], [7], [14], [15], [16], [17], [22], [25].
The LSB-based echniques, di ec ly embed he sec e da a in o he spa ial domain in an
un easonable way wi hou aking in o conside a ion he di e ence in hiding capaci y be ween
edge and smoo h a eas. In gene al, he al e a ion ole ance o an edge a ea is highe han ha o a
smoo h a ea, his meaning ha , an edge a ea can conceal mo e sec e da a han a smoo h a ea.
While human pe cep ion is less sensi i e o sub le changes in edge a eas o a pixel, i is mo e
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
3
sensi i e o change in he smoo h a eas. So, new embedding echniques using he ad an age o
edge de ec ion echnique a e in oduced, such as, [8], [10], [19] which e alua e he co ela ion
be ween neighbo ing pixels o de e mine whe he a pixel is loca ed in an edge a ea o a smoo h
a ea, and how many bi s should be embedded in ha pixel acco dingly, using me hod named
“Side Ma ch”. Ano he echnique p oposed by Wu e al. [26], using he pixel- alue di e encing
(PVD) me hod o dis inguish edge and smoo h a eas. Based on his echnique, se e al esea ches
a e sugges ed in o de o p o ide da a hiding me hods achie e a high embedding capaci y as
possible, such as, [9], [23], [24], [30].
Because he PVD me hod does no u ilize he smoo h a ea o hide la ge amoun o sec e da a, he
capaci y is s ill low. In o de o achie e highe capaci y, ano he esea che s used a combina ion
o PVD and LSB. These echniques a e based on he idea o using PVD when he di e ence
be ween a pai o pixels is la ge (edge a ea), and using LSB me hod when he di e ence is small
(smoo h a ea), such in [3], [18], [20], [27], [28], [29].
The p oposed app oach in oduces a new hyb id me hod in eg a ing bo h a dynamic classical LSB
da a hiding echnique and he pixel alue di e ence echnique. Ou p ima y a ge he e is o
inc ease he capaci y o embedded da a wi hou much dis o ion, in addi ion o inc ease he
secu i y o he p oposed scheme, key image is in oduced as he sec e key, hen ind he edges o
bo h co e and key images, hese edge pixels posi ions a e used in he embedding p ocess
consecu i ely as discussed la e .
The pape is o ganized as ollows; b ie ly backg ound abou he echniques used in he p oposed
me hod is in oduced in Sec ion 2. Ou scheme is p esen ed in Sec ion 3. The expe imen al esul s
wi h some analyses and discussions a e shown in Sec ion 4. Finally, he conclusions a e p o ided
in Sec ion 5.
2.
B
ACKGROUND
2.1 Leas Signi ican Bi Hiding (LSB) Scheme:
This me hod is one o he ea lies p oposed s eganog aphic echniques. The idea is o s o e a ixed
numbe o bi s o he sec e da a di ec ly in o he leas signi ican bi s o he pixels o he co e
image. Because LSB is ex emely simple o implemen and incu s less p ocessing ime, i is
commonly used. Howe e , he inse ion o ixed-leng h bi s in leas signi ican bi s may cause
no iceable dis o ion in he image because no all pixels can ole a e la ge changes in i s da a.
Fu he mo e, i is easy o a ack. Hence, he e is a ade-o be ween he amoun o sec e da a ha
can be embedded, he image dis o ion, and he secu i y o he s ego-image [3].
To unde s and he ope a ion o LSB eplacemen me hod, suppose we ha e he ollowing pixels,
P
1
= [11001011], P
2
= [00011010], P
3
= [01001100], and he bi s wan o embed i in he LSBs
posi ions o hem a e M = [010], he esul ed pixels a e embedding a e, P
1
= [11001010], P
2
=
[00011011], P
3
= [01001100].
So we can say ha he LSB me hod has he ollowing limi a ions [3]:
a) Since LSB is simple and well known, i becomes ulne able o secu i y a acks.
b) Inc easing he amoun o embedded da a in each pixel esul s in mo e isual deg ada ion
in he image quali y.
c) Due o he uni o m dis ibu ion o he embedded da a o e he whole co e image, he
dis up ion o he image his og am becomes no iceable.
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
4
2.2. Pixel-Value Di e encing (PVD) Scheme:
In he human isual sys em, he al e a ion o edge a eas canno be dis inguished well, unlike he
al e a ion o smoo h a eas. Wi h his concep , a no el s eganog aphy echnique was p oposed by
Wu and Tasi [26] using he pixel- alue di e encing (PVD) me hod o dis inguish edge and
smoo h a eas. This me hod elies on he idea ha no all pixels can s o e he same numbe o bi s
o he sec e da a. Ins ead o inse ing he sec e bi s di ec ly o he end o each by e o he co e
image (which is he way in which LSB wo ks), hey de e mine he numbe o bi s o be embedded
based on he di e ences be ween pai s o adjacen pixels. This allows he me hod o embed mo e
da a in he edge a ea o he co e image wi hou oo much educ ion in he s ego image quali y.
An edge is cha ac e ized by signi ican dissimila i y in g ay le els being used o indica e he
bounda y be ween wo egions in an image agmen . Edge de ec ion is a signi ican a ea o he
image p ocessing and machine ision due o he ac ha edges a e conside ed o be he impo an
ea u es o analyzing he mos essen ial in o ma ion con ained in images [11].
In he p oposed me hod, we employed he PVD idea o gene a e he edges o he co e image and
key image as well, while embedding he sec e da a.
3.
T
HE
P
ROPOSED
S
CHEME
In his sec ion, he p oposed scheme will be in oduced in de ail, i consis s o wo p ocedu es, he
embedding p ocedu e and he ex ac ing p ocedu e.
3.1 The Embedding P ocedu e:
Suppose we ha e 2 images, he co e image I
C
and he sec e key image I
K
, wi h he same size.
Any image I consis s o se o pixels:
I = {P
1
,.., P
N
},
|P
i
|= 8 bi s, 

=

,..,

, 

∈1,0. (1)
The image size is compu ed as:
N = W ×. (2)
Whe e W,  is he image wid h and heigh espec i ely. Suppose M is he sec e da a bi s, wi h
leng h n,
 = 

,

,…,

,whe e

∈1,0. (3)
The p oposed me hod akes he dependency ad an age o pixels on i s su ounding neighbo s.
The co ela ion be ween a pixel and i s neighbo s decides whe he i is loca ed in a smoo h a ea o
in an edge a ea. The da a embedding algo i hm’s s eps a e as ollows.
S ep 1: Ob ain he edge images o he co e I
C
and key I
K
g ayscale images, by di iding he
images in o o e lapping blocks, each block consis s o 4 neighbo ing pixels (Figu e 2).
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
5
P
(i ,j)
P
(i,j+1)
P
(i+1,j)
P
(i+1,j+1)
Figu e 2. A a ge pixel and h ee neighbo ing pixels.
Whe e he a ge pixel 
(,)
wi h g ay alue 
(,)
, le 
(,)
, 
(,)
and 
(,)
be he g ay
alue o he neighbo ing pixels, igh pixel as 
(,)
, down- igh pixel as 
(,)
and down
pixel as 
(,)
, espec i ely.
S ep 2: Calcula e he di e ence alue o each block which is used o ca ego ize he smoo h
and edge egions, hen selec he maximum di e ence among hem as ollowing:
1 = |
(,)
−
(,)
|,
2 = |
(,)
−
(,)
|,
3 = |
(,)
−
(,)
|,
 = $%( 1, 2, 3). (4)
S ep 3: Using Equa ion (4), we can decide whe he he a ge pixel is included in an edge a ea,
i  alue is mo e o equal o a ce ain h eshold, o he wise i is included in a smoo h a ea, a e
his s ep we will ob ain wo se s o pixels in each image, he i s called edge pixels deno ed as E,
and he o he called smoo h pixels deno ed as S,
& = &

,..,&
'
,
( =(

,..,(
)
. (5)
Whe e e, s is he size o he edge pixels se and he smoo h pixels se espec i ely,
whe e N = e + s.
S ep 4: Unselec all he in e sec ed pixels be ween he key edge pixels se &
*
and he co e edge
pixels se &
+
(.i.e. &
*
∩ &
+
), meaning he new &
*
will be (&
*
∩&
+
)
-
-
-
-
-
-
-
-
-
-
-
-
-
pixels, hen de e mine he
key edge pixel’s posi ion in he co e smoo h pixel se (
+
o ob ain he smoo h posi ions o
embedding.
S ep 5: Using he key edge pixel posi ions &
*
, selec he co esponding pixels in he smoo h a ea
o he co e image (
+
o embedding.
S ep 6: Fo embedding bi s s eam , ini ially se a a iable called a swi ch numbe (s_num)
indica es o numbe o swi ches in he embedding p ocess be ween he edge and smoo h egions.
S ep 7: To embed he sec e da a bi s, we ha e wo ca ego ies o pixels co esponding o edge
pixels ca ego y and smoo h pixels ca ego y. In each co e pixel belongs o he i s ca ego y
embeds ‘x’ sec e da a bi s using he LSBs subs i u ion echnique, o he wise embeds ‘y’ sec e
da a bi s using he LSBs subs i u ion echnique in he second ca ego y pixels.

In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
6
./(

∈ &
+
)
00 %.12
0320 00 4.12. (6)
Whe e 

is he co e pixel, &
+
is he edge pixels se o he co e image.
S ep 8: The embedding p ocess is done mu ually (no consecu i e)
be ween he selec ed pixels
o he edge &
+
and smoo h (
+
a eas. The swi ching be ween hem is de e mined acco ding o he
swi ch numbe (s_num) as de ined p e iously.
The block diag am o he embedding p ocess s eps is shown in Figu e 3.
Figu e 3. A block diag am o embedding s eps.
3.2 The Ex ac ion P ocedu e:
To eco e he o iginal sec e da a a he ecei ing side, he o iginal image 5 mus be known o
de e mine he o iginal edge E and smoo h S a eas be o e embedding, hen he ollowing s eps a e
done.
No
Coun e =0
yes
No I i > n
No
yes
S ego image
5
6
is p oduced
Di ide Co e I
C
and Key I
K
images
in o Co e
&
+
and Key
&
*
edges
Se swi ch numbe (
s_num
)
To embed M sec e bi s
Se
i=0, coun e =0
Embed x bi s in co e edge pixel


i++, coun e ++
I
coun e > s_num
yes
Embed y bi s in co e smoo h pixel


i++, coun e ++
I
coun e > s_num
Coun e =0
I i > n
No
yes
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
7
S ep 1: Find he edge and smoo h a eas o he co e and key images, as he same way desc ibed
in he p e ious sec ion.
S ep 2: Using he posi ion o he o iginal smoo h and edge a eas, we can de e mine which
pixels belong o he edge a ea E′, and which a e belong o he smoo h a ea S′ in he s ego image I′
easily.
S ep 3: Depending on swi ch numbe alue we can know he posi ion o he ce ain pixel, i i
belongs o an edge a ea o o smoo h a ea, acco ding o ha he ex ac ing o he embedded da a
bi s is done, as ollows:
./(

∈ &
6
)
0%18$91%.12
0320 0%18$914.12. (7)
S ep 4: A his s age, he e ie ing algo i hm inishes and he embedded da a has been e ie ed
comple ely. These s eps a e desc ibed in he block diag am in Figu e 4.
Figu e 4. A block diag am o ex ac ing s eps.
4.
E
XPERIMENTAL
R
ESULTS
The expe imen al esul s p esen ed in his sec ion demons a e he pe o mance o ou p oposed
scheme. To conduc ou expe imen s; we used ou 128×128 s anda d g ayscale images,
‘‘Baboon”, ‘‘Peppe ” and ‘‘Came aman” as he co e images and ‘‘Lena” as he key image,
I
coun e > s_num
I i > n
No

′
is ex ac ed
Di ide Co e I
C
and Key I
K
images
in o Co e
&
+
and Key
&
*
edges
To ex ac M sec e bi s
Se
i=0, coun e =0
I

′

∈

&
6
yes
Ex ac x bi s
Coun e ++, i++
No
yes
Ex ac y bi s
Coun e ++, i++
No
yes
I
coun e > s_num
yes
No
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
8
which a e commonly used in image p ocessing, comp ession and s eganog aphy. These images
a e shown in Figu e 5.
A se ies o pseudo andom bina y numbe s a e used as he sec e da a o be embedded in o he
co e images.
The pe o mance o he p oposed scheme is conside ed om wo iewpoin s, he isual quali y o
he s ego image and he da a capaci y.
Figu e 5. Fou 128×128 g ayscale images.
To e alua e he s ego image quali y, we used he peak signal- o-noise a io (PSNR) measu emen ,
which is used o e alua e he di e ence be ween he s ego and co e images. The (:; o mula
is de ined as
:
(:; = 10×3<
= >>
?
@AB
( C). (8)
Whe e MSE is he mean squa e e o be ween he co e and s ego images. Fo a co e image
whose wid h and heigh a e G and , (& is de ined as:
(& =

H×I
∑ ∑ K5

−5

6
L

I
M
H
M
. (9)
Whe e 5

and 5

6
a e he pixel alues o he co e and s ego images, espec i ely.
No e ha , a la ge PSNR alue means ha he s ego image is mos simila o he o iginal image
and ice e sa. Gene ally, i he PSNR alue is la ge han 30 dB, hen he dis o ion on he s ego
image is ha d o be de ec ed by human eyes [12]. Ou expe imen al esul s show ha he p oposed
scheme can embed a la ge amoun o in o ma ion while keeping an accep able isual quali y.
Second me ic o pe o mance e alua ion is he da a payload (capaci y). Da a payload can be
de ined as he amoun o in o ma ion i can hide wi hin he co e media, his can be exp essed as
numbe o bi s, which indica es he max message size ha migh be inse ed in o an image. I can
be calcula ed as a pe cen age om he ull image size [1].
In e na ional Jou nal in Founda ions o Compu e Science & Technology (IJFCST), Vol. 2, No.6, No embe 2012
9
N$O$9.14 = P
QRSTUVWX'YRZXS)'WX'[['[SRWT '
QRSTUVWX'YRZ]^'U)RWT '
_(.12/O.%03). (10)
a<1$3bc08</.1200 0 .b1<.$0 = x×no_c + y× no_k (in bi s). (11)
Whe e: no_c = numbe o co e edge pixels, no_k = numbe o key edge pixels - in e sec ed
pixels.
Usually, he high payload equi emen will con lic wi h he high PSNR equi emen . Gene ally
speaking, when he payload inc eases, he MSE will inc ease, and his will a ec he PSNR
in e sely. So, a ade-o should be made be ween capaci y and PSNR equi emen s [1]. Thus,
inno a i e da a embedding me hod p esen ed he e o imp o e he embedding payload and o
main ain he isual quali y as good as possible p ese ing he secu i y equi emen .
Table 1 shows he isual quali y o he edge images and he numbe o edge pixels in each o
hem, which a e gene a ed by ou p oposed scheme a a ce ain h eshold (..0. ≥15).
Table 1. The edge images a e gene a ed by ou scheme.
No e: *Numbe o smoo h pixels used o embedding = Numbe o key edge pixels – (key edge
pixels∩co e edgepixels).
To acili a e he compu a ion we ixed he leng h o he message o 1000 bi s, Lena image was
used as he key image. The pe o mance o he p oposed scheme is shown in Table 2. No e ha x
and y co espond o he numbe o LSBs in each edge and smoo h pixels which a e eplaced by
sec e message bi s. To main ain he quali y o he s ego image, he alue o y is se 1 o 2. The
expe imen al esul s show ha we can choose he alue o x as 3 o 4 wi hou causing any
pe cep ible dis o ion.
Ob iously, he ad an age o he p oposed scheme depends on he wo pa ame e s x and y. To
p o e ha ou scheme can achie e be e s ego image quali y and ob ain a high embedding
payload, we pe o m he p oposed algo i hm wi h a ious x and y alues. Table 2 p esen s he
ela ionship be ween hese pa ame e s and he s ego image quali y. I also show ha he la ge