)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
ANALYSIS OF NONADAPTIVE AND ADAPTIVE EDGE BASED
LSB STEGANOGRAPHY FOR COLORED IMAGES
Rohini Sharma
Deptt. of I.T,
M.M.Engg. College,
M.M.University,
Mullana (Ambala).
Ekta Walia
Deptt. of I.T,
M.M.Engg. College,
M.M.University,
Mullana (Ambala).
Deepak Sharma
Deptt. of EIE,
M.M.Engg. College,
M.M.University,
Mullana (Ambala).
Abstract
Nonadaptive and adaptive algorithms are used for edge based LSB steganography and
various papers present the results of there algorithms on grayscale images only. This
paper presents the results of analyzing the performance of nonadaptive and adaptive
edgebased LSB steganography for colored images. Both these algorithms have been
slightly modified for colored image implementation and are compared on the basis of
various evaluation parameters like peak signal noise ratio (PSNR) and mean square error
(MSE). Besides this, an algorithm has been designed and implemented to recover the
original image from the stegoimage.
Keywords: Adaptive leastsignificantbit (LSB) substitution, pixelvalue differencing
(PVD), steganography, data embedding, data extraction.
1. Introduction
Digital data communication is an essential part of everyone’s life. Data communications
have some problems such as internet security, copyright protection [1] [2] etc. To avoid
these problems, cryptography is one of the methods. However, encryption results in a
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
disordered and confusing message and can attract eavesdroppers easily. Steganography
methods [3][8] overcome this problem by hiding the secret information behind a cover
media (video, audio or image) because the presence of information can not be noticed by
any attacker. The least significant bit (LSB) substitution is a well known steganography
method [9][11]. In this method, the secret data is embedded in the cover image by
utilising LSB’s of a pixel [12]. Wang et al. proposed a method of LSB substitution to
improve the security and quality of the stegoimage [13]. Various new LSB approaches
have been proposed and some of these approaches use concept of human vision to
increase the quality of stegoimage. A “pixel value differencing” steganography method
is proposed by Wu and Tsai in 2003, that uses the difference value of two consecutive
pixels to find how many secret bits should be embedded [14]. In 2005, Wu et al.
proposed the pixel value differencing (PVD) and LSB replacement method [15]. In this
method, the two pixel values are embedded by LSB replacement method and PVD
methods, if their difference value falls into lower level and higher level respectively. If
the difference value of the two pixel changes from lower level to higher level then
readjustment of the two pixel values is required. Due to readjustment, the difference
value comes back to lower level. In 2008, Yang et al. proposed an adaptive LSB
steganographic method using pixel value differencing (PVD) for hiding data in gray
images [16]. Their method exploits the difference value of two consecutive pixels to
estimate how many secret bits will be embedded into two pixels. The pixel located in the
edge areas are embedded by k bit LSB substitution method with a larger value of k than
the pixels located in smooth areas. The range of difference values is adaptively divided
into lower, middle and higher level. The value of k is adaptive and decided by the level
which the difference value belongs to. In order to increase the quality of stegoimages, a
readjustment of the pixels values is done when embedding results cause the difference
values to fall into another range. In our paper, the analysis of nonadaptive and adaptive
steganography methods is done by comparing these methods on the basis of mean square
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
error (MSE) and peak signal to noise ratio (PSNR). The comparison of these methods has
been done for grayscale as well as color images. Besides this, the original image is
recovered from the stegoimage after the extraction of secret data.
2. NonAdaptive Edge Based LSB Steganography Method
In this algorithm, cover image is partitioned into nonoverlapping blocks of two
consecutive pixels, states
i
p and
1 + i
p . From each block we can obtain a different value
i
d by subtracting
i
p from
1 + i
p . All possible different values of
i
d ranges from 255 to
255, then
i
d ranges from 0 to 255. Therefore, the pixel
i
p and
1 + i
p is located within the
smooth area when the value
i
d is smaller and will hide less secret data. Otherwise, it is
located on the edged area and embeds more data. In smooth areas, the data is hidden into
cover image using LSB method, whereas the Pixel Value Differencing (PVD) method is
used in the edge areas. The division between the ‘lowerlevel i.e. smooth areas’ and
‘higherlevel i.e. edge areas’ of the range table is controlled by the users. Anyone who
has extracted the secret data from a stegoimage must use the original division. A
division is the key of the extracted secret data. In the lower–level of range table, each
block of two continuous pixels will hide 6bit secret data (i.e. each pixel hides 3bit secret
data), otherwise (such as the higherlevel of the range table), the bitnumber of hidden
data depends on
i
w . The data embedding algorithm is given as:
Data Embedding Procedure
First, one must assume a division Div of the ‘lowerlevel’ and ‘higherlevel’. For
example, let Div=15, so one can set the ‘lowerlevel’ to be
1
R and
2
R , ‘higherlevel’ to be
5 4 , 3
, R R R and
6
R . The detailed secret data hiding steps are as follows:
Step 1) Calculate the difference value
i
d for each block of two consecutive pixels
i
p ,
1 + i
p (for each intensity of RGB image), which is given in Eq. (1)
1 +
− =
i i i
p p d
(1)
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
Step 2) Find the optimal
i
R of the
i
d such that ( ) k u R
i i
− = min , where k u
i
≥ and
[ ]
i i i i
u l R d k , , ∈ = is the optimum
i
R for all n i ≤ ≤ 1 [15].
Step 3) If
i
R belongs to the lowerlevel, read six bits from secret data stream, and transf
orm the intensity value
i
p ,
1 + i
p into a binary value. Next, we hide 6bit secret
data into cover image by modifying
i
p and
1 + i
p according to the following
procedure. Make the 6bit secret data S =
6 5 4 3 2 1
, , , m m m m m m [15].
(1) Convert
i
p to be
'
i
p by substituting 3LSB of
i
p by
3 2 1
, m m m (for each
intensity of RGB image).
(2) Convert
1 + i
p to be
'
1 + i
p by substituting 3LSB of
1 + i
p by
6 5 4
, m m m (for each
intensity of RGB image).
(3) Calculate the new difference value
'
i
d as in Eq.(2):
'
1
' '
+
− =
i i i
p p d (2)
(4) If
'
i
d > Div (i.e. ∈
'
i
d higherlower), then readjust
'
i
p and
'
1 + i
p considering the
Eq.(3):
( )
( )
( )
' '
1
' '
1
' '
1
8, 8 ,
,
8, 8 ,
i i
i i
i i
p p
p p
p p
+
+
+
¦
− +
¦
=
´
+ −
¦
¹
if
if
' '
1
' '
1
i i
i i
p p
p p
+
+
≥
<
(3)
Step 4) If
i
R belongs to higherlevel [15]:
(1) Compute how many bits t of the secret data, which are hidden in each block
of two consecutive pixels, depend on each
i
w of the
i
R is defined
as
¸ ¸ i
w t
2
log = , where
i
w is the width of the
i
R .
(2) Read t bits binary secret data one by one according to step (1), and then
transform t into decimal valueb .
(3) Calculate the new difference value
'
d which is given by b l d
i i
+ =
'
.
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
(4) Modify the
i
p and
1 + i
p by the following formula using in Eq.(4):
( )
1
1
' '
1
1
1
, ,
2 2
, ,
2 2
,
, ,
2 2
, ,
2 2
i i
i i
i i
i i
i i
m m
p p
m m
p p
p p
m m
p p
m m
p p
+
+
+
+
+
¦
  ( (
+ −
¦ 
( (
( ¸ ¸ \ ¹
¦
¦
  ( (
¦ − +

( (
¦ ¸ ¸ ( \ ¹
=
´
  ( (
¦
− +

( (
¦
( ¸ ¸ \ ¹
¦
¦  ( (
+ −
¦ 
( (
( ¸ ¸ \ ¹
¹
if
if
if
if
'
1
'
1
'
1
'
1
&
&
&
&
i i i i
i i i i
i i i i
i i i i
p p d d
p p d d
p p d d
p p d d
+
+
+
+
≥ >
< >
≥ ≤
< ≤
(4)
Where
i i
d d m − =
'
.
After following the above steps the stego image is constructed.
Data Extraction Procedure
The following steps are executed to extract the original data.
Step 1) Partition the stegoimage into blocks of consecutive pixels and the partition proc
edure is identical with embedding [15].
Step 2) Calculate the difference value
'
i
d for each block of two consecutive pixels
( )
'
1
'
,
+ i i
p p of the stegoimage (for each intensity of RGB image), which is given in
Eq. (5)
'
1
' '
+
− =
i i i
p p d (5)
Step 3) Find the optimal
i
R of the
'
i
d according to the original range table and judge the
level of optimal
i
R denoted by the original set Div value [15].
Step 4) Extract the k LSB of the
'
i
p and
'
1 + i
p of the stegoimage directly, because the k 
LSB of the
'
i
p and
'
1 + i
p is represented by the hidden secret data [15].
After following the above steps the secret data is extracted from stego image.
3. Adaptive Edge Based LSB Steganography Method
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
The embedding strategy of the adaptive LSB substitution approach is based on the
concept that edge areas can tolerate a larger number of changes than smooth areas [16].
Similar to nonadaptive scheme, pixelvalue differencing is used to distinguish between
edge areas and smooth areas. The range [0, 255] of difference values is divided into
different levels, for instance, lower level, middle level, and higher level. For any two
consecutive pixels, both pixels are embedded by the k bit LSB substitution, but the value
k is decided by the level which their difference value belongs to. A higher level will use
a larger value of k . In order to improve the quality of the stegoimages, the wellknown
LSB substitution method, called the modified LSB substitution method is applied. The
concept of the modified LSB substitution is to increase or decrease the mostsignificant
bit (MSB) part by 1 for reducing the square error between the original pixel and the
embedded pixel. In order to extract data exactly, the difference values before and after
embedding must belong to the same level. If the difference value changes into another
level after embedding, a readjusting phase is used to readjust the pixel values. The
embedding and extracting procedures of adaptive approach are described as follows:
Data Embedding Procedure
The difference value d is computed for every nonoverlapping block with two
consecutive pixels [16]. The way of partitioning the cover image into twopixel blocks
runs through all of the rows in a raster scan. Prior to the embedding procedure, the range
[0, 255] must be divided into different levels. The first case has one dividing
line 15
12
= D , which divides the range [0, 255] into ranges [ ] 15 , 0
1
= R , [ ] 255 , 16
2
= R
where the lower level contains
1
R and the higher level contains
2
R . The second case has
two dividing lines 15
12
= D ; and 31
23
= D , which divide range [0, 255] into
ranges [ ] 15 , 0
1
= R , [ ] 31 , 16
2
= R and [ ] 255 , 32
3
= R , where
2 1
, R R and
3
R belongs to the
lower level, middle level, and higher lever, respectively. The width of range R is denoted
as R . For example, 16
1
= R , 16
2
= R and 224
3
= R . The h m l − − division means that
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
twopixel blocks with difference values falling into the lower level, middle level, and
higher level, will be embedded by the l bit, mbit and h bit LSB substitution
approaches, respectively. In addition, to succeed in the readjusting phase, we apply the
restrictions
1 2
log R l ≤ and
2 2
log R h ≤ to the h l − division, and restrictions
1 2
log R l ≤ ,
2 2
log R m ≤ and
3 2
log R h ≤ to the h m l − − division. The algorithm for the data
embedding in adaptive edge based LSB steganography method is explained as below:
Step 1) Calculate the difference value
i
d for each block with two consecutive pixels (for
each intensity of RGB image), say
i
p and
1 + i
p , using
1 +
− =
i i i
p p d .
Step 2) From the h m l − − division, find out the level to which
i
d belongs to.
Let m l k , = , and h , if
i
d belongs to the lower level, middle level, and higher
level, respectively [16].
Step 3) By the k bit LSB substitution method, embed k secret bits into
i
p and k secret
bits into
1 + i
p , respectively (for each intensity of RGB image) . Let
'
i
p and
'
1 + i
p
be the embedded results of
i
p and
1 + i
p , respectively [16].
Step 4) Apply the modified LSB substitution method to
'
i
p and
'
1 + i
p [16].
Step 5) Calculate the new difference value
'
i
d by
'
1
' '
+
− =
i i i
p p d [16].
Step 6) If
i
d and
'
i
d belong to different levels, execute the readjusting phases as follows:
case 6.1 ∈
i
d lowerlevel, ∉
'
i
d lower level. If
'
1
'
+
≥
i i
p p , readjust ( )
'
1
'
,
+ i i
p p to being
the better choice between ( )
k
i i
p p 2 ,
'
1
'
+
+
and ( )
'
1
'
, 2
+
−
i
k
i
p p ; otherwise,
readjust ( )
'
1
'
,
+ i i
p p to be the better choice between ( )
k
i i
p p 2 ,
'
1
'
−
+
and
( )
'
1
'
, 2
+
+
i
k
i
p p .
case 6.2 ∈
i
d middle level, ∈
'
i
d lower level. If
'
1
'
+
≥
i i
p p , readjust ( )
'
1
'
,
+ i i
p p to be
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
the better choice between ( )
'
1
'
, 2
+
+
i
k
i
p p and ( )
k
i i
p p 2 ,
'
1
'
−
+
; otherwise,
readjust ( )
'
1
'
,
+ i i
p p to be the better choice between ( )
k
i i
p p 2 ,
'
1
'
+
+
and
( )
'
1
'
, 2
+
−
i
k
i
p p .
case 6.3 ∈
i
d middle level, ∈
'
i
d higher level. If
'
1
'
+
≥
i i
p p , readjust ( )
'
1
'
,
+ i i
p p to be
the better choice between ( )
k
i i
p p 2 ,
'
1
'
+
+
and ( )
'
1
'
, 2
+
−
i
k
i
p p ; otherwise,
readjust ( )
'
1
'
,
+ i i
p p to be the better choice between ( )
k
i i
p p 2 ,
'
1
'
−
+
and
( )
'
1
'
, 2
+
+
i
k
i
p p .
case 6.4 ∈
i
d higher level, ∉
'
i
d higher level. If
'
1
'
+
≥
i i
p p , readjust ( )
'
1
'
,
+ i i
p p to be
the better choice between ( )
k
i i
p p 2 ,
'
1
'
−
+
and ( )
'
1
'
, 2
+
+
i
k
i
p p ; otherwise,
readjust ( )
'
1
'
,
+ i i
p p to be the better choice between( )
k
i i
p p 2 ,
'
1
'
+
+
and
( )
'
1
'
, 2
+
−
i
k
i
p p .
In Step 6), the better choice, say( )
1
,
+ i i
x x , means that it satisfies the conditions that
1 +
−
i i
x x and
i
d belong to the same level and [ ] 255 , 0
1 ,
∈
+ i i
x x , also the value of
( ) ( )
2
1 1
2
+ +
− + −
i i i i
p x p x is smaller.
After following the above steps the stego image is constructed.
Data Extraction Procedure
The stegoimage is partitioned into nonoverlapping blocks with two consecutive pixels,
and the process of extracting the embedded message is the same as the embedding
process with the same traversing order of blocks. Also, the same h m l − − division,
which is used in the embedding procedure, is used here. For each block, the detailed steps
of data extracting are as follows:
Step 1) Calculate the difference value
'
i
d for each block with two consecutive pixels (for
each intensity of RGB image), say
'
i
p and
'
1 + i
p , using
'
1
' '
+
− =
i i i
p p d .
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
Step 2) From the h m l − − division, find out the level to which
'
i
d belongs to Let m l k , = ,
and h if belongs to the lower level, middle level, and higher level, respectively.
Step 3) From the k bit LSB of a pixel, extract k secret bits from
'
i
p and k secret bits
from
'
1 + i
p [16].
After following the above steps the secret data is extracted from stego image.
4. Procedure for Recovery of Original Image from Stegoimage
The original image can be recovered from the stegoimage by using recovery procedure.
In this paper, we propose the technique to recover original image and hidden message
from stegoimage. This procedure is used for colored images. Before the embedding of
the k bits of the secret data, original k bits of the pixel values are stored at different
location. The following steps are used to obtain the original image from stegoimage:
Step 1) Calculate the difference value
i
d for each block with two consecutive pixels, say
i
p and
1 + i
p (for each intensity of RGB image), using
1 +
− =
i i i
p p d .
Step 2) From the h m l − − division, find out the level to which
i
d belongs to.
Let m l k , = , and h , if
i
d belongs to the lower level, middle level, and higher
level, respectively.
Step 3) By the k bit LSB substitution method, embed k original bits into
'
i
p and
k original bits into
'
1 + i
p , respectively. Let
i
p and
1 + i
p be the embedded results
of
'
i
p and
'
1 + i
p , respectively .
When all
'
i
p and
'
1 + i
p pixel values are replaced by
i
p and
1 + i
p , the original image is
recovered from stegoimage.
After following the above steps the cover image is reconstructed from the stego image.
5. Experimental Results
This work is implemented by using MATLAB 7.0. The objective of the work is to
compare the performance of nonadaptive and adaptive methods of steganography. The
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
methods have been implemented on set of 7 gray scale images and 7 colored images
given in Fig.1 and Fig.2 respectively. The performance of the methods have been
evaluated and compared on the basis of two measures and the measures are: Mean Square
Error (MSE) and Peak Signal to Noise Ratio (PSNR) and the equations of these two are:
( )
∑∑
−
=
−
=
−
×
=
1
1
1
1
2
ˆ
1
m
i
n
j
ij ij
I I
n m
MSE (6)
PSNR =
( )
(
¸
(
¸
MSE
2
10
255
log 10 (7)
(a)
(b)
(c)
(d)
(e)
(f)
(g)
Fig.1 Grayscale images. (a) Baboon (b) Boat (c) Cameraman (d) Jetplane (e) Peppers
(f) Pirate (g) Tree
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
(a)
(b)
(c)
(d)
(e)
(f)
(g)
Fig.2 Colored images. (a) Baboon (b) Boat (c) Goldhill (d) House (e) Lenna
(f) Masuda (g) Peppers
These measures are calculated for grayscale as well as colored images. The following
equations are used to compute them: The table 1 shows the Mean Square Error (MSE)
and Peak Signal to Noise Ratio (PSNR) of grayscale images for both methods given as:
Table 1 MSE and PSNR for NonAdaptive & Adaptive method for grayscale images
NonAdaptive
(GrayScale)
Adaptive
(GrayScale)
S.No
Images
(512 X 512 pixels)
MSE PSNR MSE PSNR
1 Baboon 4.75 41.39 4.72 41.42
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
2 Boat 4.77 41.37 4.73 41.41
3 Cameraman 4.72 41.42 4.70 41.44
4 Jetpalne 4.66 41.48 4.63 41.50
5 Peppers 4.67 41.47 4.66 41.48
6 Pirate 4.81 41.34 4.81 41.34
7 Tree 4.91 41.25 4.89 41.26
Average 4.76 41.38 4.73 41.41
The MSE and PSNR are calculated for more than one image that’s why the average value
of MSE and PSNR is also shown in the table 1. Table 2 shows the MSE and PSNR of
colored images for both methods. The table of the MSE and PSNR for colored images is
given as:
Table 2 MSE and PSNR for NonAdaptive & Adaptive method for colored images
NonAdaptive
(Colored)
Adaptive
(Colored)
S.No Images
(512 X 512 pixels)
MSE PSNR MSE PSNR
1 Baboon 4.78 41.33 4.79 41.32
2 Boat 5.84 40.46 8.27 38.95
3 Goldhill 5.77 40.52 9.06 38.56
4 House 4.66 41.45 4.67 41.44
5 Lenna 4.71 41.39 4.75 41.36
6 Masuda 4.33 41.76 4.34 41.75
7 Peppers 4.61 41.49 4.64 41.46
Average 4.95 41.20 5.79 40.69
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
0 2 4 6 8
4.6
4.65
4.7
4.75
4.8
4.85
4.9
4.95
Image No.
M
e
a
n
s
q
u
a
r
e
E
r
r
o
r
(
M
S
E
)
Grayscale Images
0 2 4 6 8
4
5
6
7
8
9
10
Image No.
M
e
a
n
s
q
u
a
r
e
E
r
r
o
r
(
M
S
E
)
Colored Images
NonAdaptive
Adaptive
NonAdaptive
Adaptive
Fig. 3(a) Fig. 3(b)
Fig. 3(a) Comparison of MSE of nonadaptive and adaptive method (grayscale),
Fig. 3(b) Comparison of MSE of nonadaptive and adaptive method (colored)
The comparisons of nonadaptive and adaptive steganography methods on the basis of
MSE for grayscale as well as colored images are shown in Fig.3 (a) and (b) and for
PSNR comparisons are shown in Fig. 4 (a) and (b) respectively. The MSE in case of
grayscale image is more and the PSNR is less for nonadaptive steganography method
than adaptive steganography method. The less MSE and more PSNR is the desirable
condition for better performance. The adaptive steganography method satisfies this so,
adaptive method is better choice for grayscale images. For colored images the MSE is
less and PSNR is more in nonadaptive steganography method than in adaptive
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
steganography method. As compared to adaptive method, the performance of non
adaptive method is better for colored images.
0 2 4 6 8
41.25
41.3
41.35
41.4
41.45
41.5
Image No.
P
e
a
k
S
i
g
n
a
l
N
o
i
s
e
R
a
t
i
o
(
P
S
N
R
)
Grayscale Images
0 2 4 6 8
38.5
39
39.5
40
40.5
41
41.5
42
Image No.
P
e
a
k
S
i
g
n
a
l
N
o
i
s
e
R
a
t
i
o
(
P
S
N
R
)
Colored Images
Nonadaptive
Adaptive
Nonadaptive
Adaptive
Fig. 4(a) Fig. 4(b)
Fig. 4(a) Comparison of PSNR of nonadaptive and adaptive method (grayscale),
Fig. 4(b) Comparison of PSNR of nonadaptive and adaptive method (colored)
5. Conclusions
In this paper, performance of adaptive and nonadaptive methods is analyzed for edge
based LSB steganography. The image quality after data embedding is very important for
better performance of steganography methods. The image quality is evaluated by Mean
Square Error (MSE) and Peak Signal to Noise ratio (PSNR) for nonadaptive and
adaptive methods for grayscale as well as colored images. The following conclusions
can be made from this analysis:
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
1) For grayscale images, a trend is observed that PSNR is higher and MSE is less in case
of adaptive method, therefore adaptive method is better for grayscale images.
2) For colored images, a trend is observed that PSNR is higher and MSE is less in case of
nonadaptive method, therefore nonadaptive is better for colored images.
3) 100% recovery of original image is observed after extraction of secret data.
6. References
[1] D. Zou, C. W.Wu, G. Xuan, and Y. Q. Shi, “A ContentBased Authentication System
with Lossless Data Hiding,” in Proc. Multimedia Expo Int. Conf., vol. 2, pp. 6–9,
2003.
[2] T. Liu and Z. D. Qiu, “The Survey of Digital WatermarkingBased Image
Authentication Techniques,” in Signal Process., vol. 2, pp. 1556–1559, 2002.
[3] D. Artz, “Digital Steganographic: Hiding Data within Data,” IEEE Internet Comput.,
vol. 5, no. 3, pp. 75–80, May/Jun. 2001.
[4] H. J. Highland, “Data Encryption: A NonMathematical Approach,” Comput.
Security, vol. 16, pp. 369–386, 1997.
[5] Y. H. Chu and S. Chang, “Dynamical Cryptography Based on Synchronized Chaotic
Systems,” Inst. Elect. Eng. Electron. Lett., vol. 35, no. 12, pp. 974–975, 1999.
[6] D. W. Bender, N. M. Gruhl, and A. Lu, “Techniques for Data Hiding,” IBM Syst. J.,
vol. 35, pp. 313–316, 1996.
[7] R. R. Anderson and F. A. P. Peticolas, “On The Limits of Steganography,” IEEE J.
Sel. Areas Commun., vol. 16, no. 4, pp. 474–481, May 1998.
[8] Y. K. Lee and L. H. Chen, “High Capacity Image Steganography Model,” Proc. Inst.
Elect. Eng. Vis. Image, Signal Processing, vol. 147, no. 3, pp. 288–294, 2000.
[9] R. Z. Wang, C. F. Lin, and J. C. Lin, “Image Hiding by Optimal LSB Substitution and
Genetic Algorithm,” Pattern Recognit., vol. 34, no. 3, pp. 671–683, 2000.
[10] C. C. Thien and J. C. Lin, “A Simple and HighHiding Capacity Method for Hiding
DigitByDigit Data in Images Based on Modulus Function,” Pattern Recognit., vol.
36, no. 3, pp. 2875–2881, 2003.
International Journal of Computing and Business Research
(IJCBR)
ISSN (Online) : 22296166
Volume 2 Issue 1 2011
[11] C. K. Chan and L. M. Chen, “Hiding Data in Images by Simple LSB Substitution,”
Pattern Recognit., vol. 37, no. 3, pp. 469–474, 2004.
[12] Walton, S., “Image Authentication for a Slippery New Age”, Dr. Dobb’s J., vol. 20,
no. 4, pp. 18–26, 1995,.
[13] Wang, R.Z., Lin, C.F., and Lin, J.C., “Image Hiding by Optimal LSB Substitution
and Genetic Algorithm”, Pattern Recognit., vol. 34, no. 3, pp. 671–683, 2001.
[14] D. C. Wu and W. H. Tsai, “A Steganographic Method for Images by PixelValue
Differencing,” Pattern Recognit. Lett., vol. 24, no. 9–10, pp. 1613–1626, 2003.
[15] H. C. Wu, N. I. Wu, C. S. Tsai, and M. S. Hwang, “Image Steganographic Scheme
Based on PixelValue Differencing and LSB Replacement Methods,” Proc. Inst.
Elect. Eng., Vis. Images Signal Process., vol. 152, no. 5, pp. 611–615, 2005.
[16] ChengHsing Yang, ChiYao Weng, ShiuhJeng Wang, “Adaptive Data Hiding in
Edge Areas of Images with Spatial LSB Domain Systems.” IEEE Transactions on.
Information Forensics and Security, Vol. 3, no. 3, September 2008.
[17] Rafel C. Gonzalez, Richard E. Woods, “Digital Image Processing” 2
nd
ed., 2002.
if their difference value falls into lower level and higher level respectively. proposed the pixel value differencing (PVD) and LSB replacement method [15]. If the difference value of the two pixel changes from lower level to higher level then readjustment of the two pixel values is required. Various new LSB approaches have been proposed and some of these approaches use concept of human vision to increase the quality of stegoimage. The pixel located in the edge areas are embedded by k bit LSB substitution method with a larger value of k than the pixels located in smooth areas. A “pixel value differencing” steganography method is proposed by Wu and Tsai in 2003. the analysis of nonadaptive and adaptive steganography methods is done by comparing these methods on the basis of mean square . The range of difference values is adaptively divided into lower. Wang et al.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 disordered and confusing message and can attract eavesdroppers easily. Yang et al. the secret data is embedded in the cover image by utilising LSB’s of a pixel [12]. The least significant bit (LSB) substitution is a well known steganography method [9][11]. audio or image) because the presence of information can not be noticed by any attacker. a readjustment of the pixels values is done when embedding results cause the difference values to fall into another range. that uses the difference value of two consecutive pixels to find how many secret bits should be embedded [14]. In our paper. Steganography methods [3][8] overcome this problem by hiding the secret information behind a cover media (video. In this method. In this method. In order to increase the quality of stegoimages. In 2008. middle and higher level. Their method exploits the difference value of two consecutive pixels to estimate how many secret bits will be embedded into two pixels. The value of k is adaptive and decided by the level which the difference value belongs to. Wu et al. the difference value comes back to lower level. proposed a method of LSB substitution to improve the security and quality of the stegoimage [13]. proposed an adaptive LSB steganographic method using pixel value differencing (PVD) for hiding data in gray images [16]. In 2005. Due to readjustment. the two pixel values are embedded by LSB replacement method and PVD methods.
it is located on the edged area and embeds more data. The division between the ‘lowerlevel i. 2. cover image is partitioned into nonoverlapping blocks of two consecutive pixels. each block of two continuous pixels will hide 6bit secret data (i. All possible different values of d i ranges from 255 to 255. ‘higherlevel’ to be R3. which is given in Eq. R5 and R6 . states pi and pi +1 .International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 error (MSE) and peak signal to noise ratio (PSNR). let Div=15. so one can set the ‘lowerlevel’ to be R1 and R2 . smooth areas’ and ‘higherlevel i. the data is hidden into cover image using LSB method. (1) d i = p i − p i +1 (1) . Besides this. one must assume a division Div of the ‘lowerlevel’ and ‘higherlevel’. The data embedding algorithm is given as: Data Embedding Procedure First. pi +1 (for each intensity of RGB image). In smooth areas. Anyone who has extracted the secret data from a stegoimage must use the original division.e. For example. The detailed secret data hiding steps are as follows: Step 1) Calculate the difference value d i for each block of two consecutive pixels pi . the pixel pi and pi +1 is located within the smooth area when the value di is smaller and will hide less secret data. R4 .e.e. otherwise (such as the higherlevel of the range table). Therefore. edge areas’ of the range table is controlled by the users. Otherwise. A division is the key of the extracted secret data. The comparison of these methods has been done for grayscale as well as color images. the original image is recovered from the stegoimage after the extraction of secret data. NonAdaptive Edge Based LSB Steganography Method In this algorithm. In the lower–level of range table. the bitnumber of hidden data depends on wi . From each block we can obtain a different value d i by subtracting pi from pi +1 . then di ranges from 0 to 255. each pixel hides 3bit secret data). whereas the Pixel Value Differencing (PVD) method is used in the edge areas.
and transform the intensity value pi .(3): (p . ( ( ) ) if pi' ≥ pi' +1 if pi' < pi' +1 (3) Step 4) If Ri belongs to higherlevel [15]: (1) Compute how many bits t of the secret data.p ) ' i ' i +1 pi' − 8.(2): d i' = p i' − p i' +1 (2) (4) If d i' > Div (i. then readjust pi' and pi' +1 considering the Eq. m6 [15]. (2) Read t bits binary secret data one by one according to step (1). pi' +1 + 8 . where u i ≥ k and k = d i . (2) Convert pi +1 to be pi' +1 by substituting 3LSB of pi +1 by m4 m5 . (3) Calculate the new difference value d ' which is given by d i' = l i + b . pi +1 into a binary value. m6 (for each intensity of RGB image). m2 m3 (for each intensity of RGB image). where wi is the width of the Ri .International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 Step 2) Find the optimal Ri of the d i such that Ri = min(u i − k ) . Ri ∈ [l i . (3) Calculate the new difference value d i' as in Eq. Step 3) If Ri belongs to the lowerlevel. which are hidden in each block of two consecutive pixels.e. Make the 6bit secret data S = m1 . . = ' ' pi + 8. pi +1 − 8 . m2 m3 m4 . m5 . Next. and then transform t into decimal value b . we hide 6bit secret data into cover image by modifying pi and pi +1 according to the following procedure. d i' ∈ higherlower). (1) Convert pi to be pi' by substituting 3LSB of pi by m1 . read six bits from secret data stream. depend on each wi of the Ri is defined as t = log 2 wi . u i ] is the optimum Ri for all 1 ≤ i ≤ n [15].
Adaptive Edge Based LSB Steganography Method . (5) d i' = p i' − p i' +1 (5) Step 3) Find the optimal Ri of the d i' according to the original range table and judge the level of optimal Ri denoted by the original set Div value [15]. which is given in ' i ' i +1 Eq. pi +1 − . because the k LSB of the pi' and pi' +1 is represented by the hidden secret data [15]. After following the above steps the secret data is extracted from stego image. p ) of the stegoimage (for each intensity of RGB image). After following the above steps the stego image is constructed. 2 2 m m pi − . pi +1 − . pi +1 + . 2 2 m m pi + . 3. Step 4) Extract the k LSB of the pi' and pi' +1 of the stegoimage directly.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 (4) Modify the pi and pi +1 by the following formula using in Eq.p ) ' i ' i +1 ' if p i < p i +1 & d i > d i (4) if p i ≥ p i +1 & d i' ≤ d i if p < p & d ' ≤ d i i +1 i i Where m = d i' − d i . Data Extraction Procedure The following steps are executed to extract the original data. 2 2 ' if p i ≥ p i +1 & d i > d i (p. pi +1 + . Step 1) Partition the stegoimage into blocks of consecutive pixels and the partition procedure is identical with embedding [15]. 2 2 m m pi − .(4): = m m pi + . Step 2) Calculate the difference value d i' for each block of two consecutive pixels ( p .
which divides the range [0. The second case has two dividing lines D12 = 15 . The width of range R is denoted as R . respectively. which divide range [0. For example. A higher level will use a larger value of k . R2 = 16 and R3 = 224 . middle level. and D23 = 31 . lower level. 255] into ranges R1 = [0. The embedding and extracting procedures of adaptive approach are described as follows: Data Embedding Procedure The difference value d is computed for every nonoverlapping block with two consecutive pixels [16]. R2 = [16. R1 = 16 . for instance. The concept of the modified LSB substitution is to increase or decrease the mostsignificantbit (MSB) part by 1 for reducing the square error between the original pixel and the embedded pixel. Prior to the embedding procedure. and higher level.15] .15] . but the value k is decided by the level which their difference value belongs to. If the difference value changes into another level after embedding. the wellknown LSB substitution method. where R1 .255] . 255] of difference values is divided into different levels. Similar to nonadaptive scheme. the range [0. and higher lever. 255] into ranges R1 = [0. a readjusting phase is used to readjust the pixel values. 255] must be divided into different levels. middle level. R2 = [16.255] where the lower level contains R1 and the higher level contains R2 . The first case has one dividing line D12 = 15 . For any two consecutive pixels. In order to extract data exactly.31] and R3 = [32. The way of partitioning the cover image into twopixel blocks runs through all of the rows in a raster scan. pixelvalue differencing is used to distinguish between edge areas and smooth areas. In order to improve the quality of the stegoimages. The l − m − h division means that .International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 The embedding strategy of the adaptive LSB substitution approach is based on the concept that edge areas can tolerate a larger number of changes than smooth areas [16]. the difference values before and after embedding must belong to the same level. The range [0. called the modified LSB substitution method is applied. both pixels are embedded by the k bit LSB substitution. R2 and R3 belongs to the lower level.
Step 3) By the k bit LSB substitution method. and higher level. m ≤ log 2 R2 and h ≤ log 2 R3 to the l − m − h division. middle level. pi' +1 + 2 k and pi' − 2 k . Step 4) Apply the modified LSB substitution method to pi' and pi' +1 [16]. The algorithm for the data embedding in adaptive edge based LSB steganography method is explained as below: Step 1) Calculate the difference value d i for each block with two consecutive pixels (for each intensity of RGB image). respectively [16]. Step 2) From the l − m − h division. p ) to be readjust ' i ' i ' i +1 ' i ' i +1 k k ' i +1 i ' i ' i ' i +1 ' i ' i +1 . readjust ( p . respectively (for each intensity of RGB image) . p ) to be the better choice between ( p . using d i = pi − pi +1 . execute the readjusting phases as follows: case 6. if d i belongs to the lower level. we apply the restrictions l ≤ log 2 R1 and h ≤ log 2 R2 to the l − h division. Step 6) If d i and d i' belong to different levels. pi' +1 to being the better choice between pi' . case 6. p − 2 ) and ( p + 2 . will be embedded by the l bit. In addition. embed k secret bits into pi and k secret bits into pi +1 .1 d i ∈ lowerlevel. Step 5) Calculate the new difference value d i' by d i' = pi' − pi' +1 [16]. and higher level. ( ) ( ) ( ) ( p . say pi and pi +1 . d ∈ lower level. d i' ∉ lower level.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 twopixel blocks with difference values falling into the lower level. If p ≥ p . to succeed in the readjusting phase. otherwise. respectively [16]. m . pi' +1 . respectively. and restrictions l ≤ log 2 R1 . readjust pi' . middle level. Let pi' and pi' +1 be the embedded results of pi and pi +1 . find out the level to which d i belongs to. and h .2 d ∈ middle level. m bit and h bit LSB substitution approaches. Let k = l . If pi' ≥ pi' +1 . p ).
d ∉ higher level. say pi' and pi' +1 . pi' +1 and pi' . If p ≥ p . p + 2 ) and (p − 2 .4 d ∈ higher level. case 6. otherwise. readjust ( p . p ) to be the better choice between ( p . xi +1 ) .International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 the better choice between pi' + 2 k . the detailed steps of data extracting are as follows: Step 1) Calculate the difference value d i' for each block with two consecutive pixels (for each intensity of RGB image). p ) to be the better choice between ( p . After following the above steps the stego image is constructed. p ). which is used in the embedding procedure. For each block. Data Extraction Procedure The stegoimage is partitioned into nonoverlapping blocks with two consecutive pixels. d ∈ higher level. is used here. means that it satisfies the conditions that xi − xi +1 and d i belong to the same level and xi . Also. p ). otherwise. xi +1 ∈ [0. p ) . and the process of extracting the embedded message is the same as the embedding process with the same traversing order of blocks. p − 2 ) and ( p + 2 . If p ≥ p . p ) . ( ) ( ) ( p . also the value of (xi − pi )2 + (xi +1 − pi +1 )2 is smaller. readjust ( p . the same l − m − h division. p ) to be the better choice between ( p . p ) to be the better choice between ( p . p + 2 ) and (p − 2 . . p )to be the better choice between ( p .255].3 d ∈ middle level. pi' +1 − 2 k . readjust ' i k ' i ' i +1 ' i ' i +1 k ' i +1 i ' i ' i ' i +1 ' i ' i +1 ' i ' i +1 k ' i k ' i +1 ' i ' i +1 ' i ' i +1 k ' i k ' i +1 i ' i ' i ' i +1 ' i ' i +1 ' i ' i +1 k ' i k ' i +1 ' i ' i +1 ' i ' i +1 k ' i k ' i +1 In Step 6). readjust ( p . the better choice. readjust ( p . otherwise. say ( xi . p + 2 ) and ( p − 2 . case 6. p − 2 ) and ( p + 2 . p ) . using d i' = pi' − pi' +1 .
middle level. After following the above steps the secret data is extracted from stego image. Step 3) From the k bit LSB of a pixel. m . This procedure is used for colored images. 5. The objective of the work is to compare the performance of nonadaptive and adaptive methods of steganography. and higher level. embed k original bits into pi' and k original bits into pi' +1 . Let k = l . and h . find out the level to which d i' belongs to Let k = l .International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 Step 2) From the l − m − h division. Before the embedding of the k bits of the secret data. After following the above steps the cover image is reconstructed from the stego image. In this paper. find out the level to which d i belongs to. The . and h if belongs to the lower level. original k bits of the pixel values are stored at different location. the original image is recovered from stegoimage. When all pi' and pi' +1 pixel values are replaced by pi and pi +1 . Let pi and pi +1 be the embedded results of pi' and pi' +1 . m . extract k secret bits from pi' and k secret bits from pi' +1 [16]. 4. middle level. and higher level. respectively. using d i = pi − pi +1 . if d i belongs to the lower level. Procedure for Recovery of Original Image from Stegoimage The original image can be recovered from the stegoimage by using recovery procedure. we propose the technique to recover original image and hidden message from stegoimage. Experimental Results This work is implemented by using MATLAB 7. respectively . respectively. Step 2) From the l − m − h division. Step 3) By the k bit LSB substitution method. say pi and pi +1 (for each intensity of RGB image). The following steps are used to obtain the original image from stegoimage: Step 1) Calculate the difference value d i for each block with two consecutive pixels. respectively.0.
International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 methods have been implemented on set of 7 gray scale images and 7 colored images given in Fig. (a) Baboon (b) Boat (c) Cameraman (d) Jetplane (e) Peppers (f) Pirate (g) Tree .1 Grayscale images.1 and Fig. The performance of the methods have been evaluated and compared on the basis of two measures and the measures are: Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR) and the equations of these two are: MSE = 1 m−1 n−1 ∑∑ I ij − Iˆij m × n i =1 j =1 ( ) 2 (6) (255)2 PSNR = 10 log10 MSE (7) (a) (b) (c) (d) (e) (f) (g) Fig.2 respectively.
No (512 X 512 pixels) NonAdaptive (GrayScale) MSE PSNR 41.42 1 Baboon 4.75 .2 Colored images.72 PSNR 41. (a) Baboon (b) Boat (c) Goldhill (d) House (e) Lenna (f) Masuda (g) Peppers These measures are calculated for grayscale as well as colored images.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 (a) (b) (c) (d) (e) (f) (g) Fig.39 Adaptive (GrayScale) MSE 4. The following equations are used to compute them: The table 1 shows the Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR) of grayscale images for both methods given as: Table 1 MSE and PSNR for NonAdaptive & Adaptive method for grayscale images Images S.
73 4.37 41.33 40.72 4.20 Adaptive (Colored) MSE 4. Table 2 shows the MSE and PSNR of colored images for both methods.63 4.84 5.76 41.45 41.33 4.66 4.41 Average 4.25 41.No Images (512 X 512 pixels) NonAdaptive (Colored) MSE PSNR 41.26 41.95 Average .34 41.48 41.50 41.46 40.47 41.44 41.56 41.64 5. The table of the MSE and PSNR for colored images is given as: Table 2 MSE and PSNR for NonAdaptive & Adaptive method for colored images S.89 4.44 41.67 4.52 41.27 9.75 41.76 The MSE and PSNR are calculated for more than one image that’s why the average value of MSE and PSNR is also shown in the table 1.79 8.81 4.69 1 2 3 4 5 6 7 Baboon Boat Goldhill House Lenna Masuda Peppers 4.34 41.73 41.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 2 3 4 5 6 7 Boat Cameraman Jetpalne Peppers Pirate Tree 4.71 4.38 4.91 41.66 4.77 4.41 41.06 4.48 41.34 4.75 4.77 4.66 4.61 4.32 38.36 41.95 38.70 4.39 41.79 PSNR 41.81 4.78 5.42 41.67 4.49 41.46 40.
7 4. 3(b) Fig. 3(a) Comparison of MSE of nonadaptive and adaptive method (grayscale).75 6 4.6 4 0 2 4 Image No. 6 8 4. 3(a) Fig.65 5 4. 3(b) Comparison of MSE of nonadaptive and adaptive method (colored) The comparisons of nonadaptive and adaptive steganography methods on the basis of MSE for grayscale as well as colored images are shown in Fig.3 (a) and (b) and for PSNR comparisons are shown in Fig. The adaptive steganography method satisfies this so.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 Grayscale Images 4. 6 8 Fig. The MSE in case of grayscale image is more and the PSNR is less for nonadaptive steganography method than adaptive steganography method.8 7 4.85 8 4. 4 (a) and (b) respectively.9 9 10 Colored Images NonAdaptive Adaptive Mean square Error (MSE) Mean square Error (MSE) 0 2 4 Image No. Fig. The less MSE and more PSNR is the desirable condition for better performance. For colored images the MSE is less and PSNR is more in nonadaptive steganography method than in adaptive .95 NonAdaptive Adaptive 4. adaptive method is better choice for grayscale images.
International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 steganography method. performance of adaptive and nonadaptive methods is analyzed for edge based LSB steganography. 4(a) Fig.5 41. Conclusions In this paper. 6 8 Fig. 6 8 38. the performance of nonadaptive method is better for colored images.5 0 2 4 Image No.4 40. 4(a) Comparison of PSNR of nonadaptive and adaptive method (grayscale).25 0 2 4 Image No.45 Peak Signal Noise Ratio (PSNR) Peak Signal Noise Ratio (PSNR) 41 Colored Images 41. The following conclusions can be made from this analysis: .3 39 Nonadaptive Adaptive 41. The image quality after data embedding is very important for better performance of steganography methods. The image quality is evaluated by Mean Square Error (MSE) and Peak Signal to Noise ratio (PSNR) for nonadaptive and adaptive methods for grayscale as well as colored images. As compared to adaptive method. 4(b) Comparison of PSNR of nonadaptive and adaptive method (colored) 5. 4(b) Fig.35 40 39. Fig.5 42 Nonadaptive Adaptive 41.5 41. Grayscale Images 41.5 41.
F. and J. pp. “Techniques for Data Hiding. [7] R. N. vol. therefore nonadaptive is better for colored images. 35. no. pp. P. pp. 36.. 2) For colored images.. “Data Encryption: A NonMathematical Approach. 3. [2] T. 3) 100% recovery of original image is observed after extraction of secret data. Sel. Lett. Anderson and F. Z. pp. 2001. Peticolas. vol. G. no.” IEEE Internet Comput. C. vol. 5. no. 147. Xuan. “High Capacity Image Steganography Model. 474–481. 1999. Lin. vol. Q.. 1556–1559. 6. 974–975. Areas Commun. H. R. 16. 1997. “A ContentBased Authentication System with Lossless Data Hiding.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 1) For grayscale images. Elect. 16.. W. W. a trend is observed that PSNR is higher and MSE is less in case of adaptive method. Electron.. [10] C. 671–683. no. pp. vol.” IBM Syst. [8] Y.. Lee and L. C. Lin.” Inst. vol. 2000. References [1] D. Security. Chen. vol. Shi. “A Simple and HighHiding Capacity Method for Hiding DigitByDigit Data in Images Based on Modulus Function.” Pattern Recognit. Gruhl. no. 75–80. C. 1996. and A. “On The Limits of Steganography. 2003. therefore adaptive method is better for grayscale images.” in Signal Process. 12. [5] Y..” Pattern Recognit. Wang. 4. 6–9. Thien and J. 288–294. M. no. K. [3] D. pp.” IEEE J. May/Jun.” Proc. 2000. [6] D. 34. “The Survey of Digital WatermarkingBased Image Authentication Techniques. Chang. “Dynamical Cryptography Based on Synchronized Chaotic Systems.. Highland. [9] R. A. 2875–2881. Liu and Z. C. pp. and Y. H. 313–316. Lin. J. a trend is observed that PSNR is higher and MSE is less in case of nonadaptive method. “Digital Steganographic: Hiding Data within Data. Eng. “Image Hiding by Optimal LSB Substitution and Genetic Algorithm. 2. vol. 369–386. J. vol. Vis. pp. Artz. 3. [4] H. 3. 2003. pp. Inst. Image. Eng. C. D. Elect. Signal Processing. . Multimedia Expo Int. Chu and S. Lu.” in Proc. vol. Qiu.” Comput. 3.Wu. 2002. 2. Conf. Zou. Bender. May 1998. 35. pp.
Lin. Vis. S. 4. S. ChiYao Weng. .” Proc. C.. 2002. “Digital Image Processing” 2nd ed. 3. no. Wu and W. Woods. 18–26. [13] Wang. “Image Hiding by Optimal LSB Substitution and Genetic Algorithm”.. Pattern Recognit. 2001. Wu. 2004. 37. Eng.. Hwang.. 20. Gonzalez. Images Signal Process. H. 2003. pp. 611–615. [14] D. 3. Elect. Inst. Vol. no. no. C.” IEEE Transactions on. Tsai. 5. no. 1613–1626. Tsai.” Pattern Recognit. no.Z. vol. 3. 3. J. “Adaptive Data Hiding in Edge Areas of Images with Spatial LSB Domain Systems. N. R. Chen.. Richard E. vol.. Wu.” Pattern Recognit. vol. 34.. Chan and L. and M. Dr.International Journal of Computing and Business Research (IJCBR) ISSN (Online) : 22296166 Volume 2 Issue 1 2011 [11] C. 671–683. Dobb’s J. vol. C. pp.C. [16] ChengHsing Yang. 469–474. [15] H. 1995. 24. K. vol. Lett. Information Forensics and Security. pp.. [12] Walton. ShiuhJeng Wang. [17] Rafel C.. pp. pp... and Lin. “Image Steganographic Scheme Based on PixelValue Differencing and LSB Replacement Methods.F. “A Steganographic Method for Images by PixelValue Differencing. “Hiding Data in Images by Simple LSB Substitution. September 2008. no. 152. “Image Authentication for a Slippery New Age”. 9–10. M. 2005. S. I. C..