Steganography is the science that involves
communicating secret information in an appropriate
carrier so no one apart from the sender and the recipient
even can recognize that there is hidden
information. Steganography is the art of hiding
messages inside unsuspicious medium such as images,
videos, various types of files…etc. It's a method to
establish a secure communication channel between two
parties. The purpose of steganography is to hide the
existence of a message from an eavesdropper or third
parties. Steganalysis is the branch of data processing
that seeks the identification of carrier vessels and
retrieval of message hidden. In this paper we present
enhanced implementation for Steganography algorithm,
an algorithm that we claim to be safe, built over DCT
(Discrete Cosine Transformation) frequency
domain mutation[12], the algorithm uses error reductive
measurements such as pattern matching to obtain
a reasonable a better image quality by reducing number
of changes that steganography algorithm made during
the embedding process.
GenAI talk for Young at Wageningen University & Research (WUR) March 2024
37 c 551 - reduced changes in the carrier of steganography algorithm
1. Reduced Changes in the Carrier of Steganography Algorithm
Mohammed Kharma 1
, Dr. Nedal Kafri 2
1: Al Quds University(Moh.kharma@gmail.com), 2: Al Quds University(nkafri@science.alquds.edu)
ABSTRACT
Steganography is the science that involves
communicating secret information in an appropriate
carrier so no one apart from the sender and the recipient
even can recognize that there is hidden
information. Steganography is the art of hiding
messages inside unsuspicious medium such as images,
videos, various types of files…etc. It's a method to
establish a secure communication channel between two
parties. The purpose of steganography is to hide the
existence of a message from an eavesdropper or third
parties. Steganalysis is the branch of data processing
that seeks the identification of carrier vessels and
retrieval of message hidden. In this paper we present
enhanced implementation for Steganography algorithm,
an algorithm that we claim to be safe, built over DCT
(Discrete Cosine Transformation) frequency
domain mutation[12], the algorithm uses error reductive
measurements such as pattern matching to obtain
a reasonable a better image quality by reducing number
of changes that steganography algorithm made during
the embedding process.
I.INTRODUCTION
Information security and hiding is a general term
involves several sub disciplines and areas around a
wide spectrum of problems like embedding message
into other contents. Information hiding concept denotes
to maintain the confidentiality of information or making
the information cannot be detected. Many techniques’
have been developed to hide the secret information in
other container data to be viewed as an innocent.
Steganography is the art and science of writing
hidden messages inside innocent looking containers in
such a way that no one apart from the sender and
intended recipient even realizes the existence of a
hidden message. Steganography differs from
cryptography in that the first makes the message
unreadable while the second makes it unseen. It is
nevertheless possible to use both techniques to add
security to the messages.
Steganography is a two-part word of Greek origin.
“Stegano graphy” or “cover/hidden/roof writing”[19].
Its ancient origins can be traced back to 440 BC When
Demeratus sent a warning about a forthcoming attack to
Greece by writing it on a wooden panel and covering it
in wax. A second classic example is that of Histiaeus,
who had shaved the head of his most trusted slave and
tattooed a message on it. After his hair had grown the
message was hidden. The purpose was to instigate a
revolt against the Persians. The third classic example is
to hide the message throughout the first character in the
subject paragraph, so construction of the message can be
achieved by taking the first letter from every paragraph.
Steganography used in electronic communication
include steganographic coding inside of a transport
layer, such as an MP3 file, or a protocol, such as TCP
and UDP. A wide variety of steganography
implementations in sound files, movies, exe files, videos
and many other exiting file types. The technique have
had a lot of attention after the USA government had
claimed the technique used by al-Quada terrorists in
there communication, Claims that were afterwards
proven to be false [1].
In secure communication model and to illustrate
steganography problem, participated parties in
the communication can be summarized as: Alice and
Bob are trying to communicate a secure message.
However, all there communications are being checked
and filtered by third party who want to know the secrete
message they want to communicate, to achieve the non
delectability in Alice and Bob communication so they
use another container called cover-object to embed the
secrete message into it[1].
Discrete cosine transform (DCT) is the most well-
known transform coding techniques for converting a
signal into elementary frequency components used to
implement lossy image compression (such as JPEG
format) to transform the image from the spatial domain
to the frequency domain. DCT separates the image into
three different frequency components: high, medium
and low where the image is segmented into non-overlap
8 pixels x 8 pixels blocks. The DCT is computed for
each block starting from left corner to right corner in
top-bottom order [13].
The rest of this paper is organized as follows:
Section 2 gives a background regarding the main
schemes of steganography; spatial domain and
frequency domain, and their evaluation techniques.
While Section 3 introduces the proposed steganography
2. algorithm Section 4 presents and discusses the obtained
experimental results. Finally, Section 5 concludes the
paper.
II. BACKGROUND
Steganography is one of main aspects of secure
communication channels and widely used techniques
that manipulate information in order to hide their
existence rather than encrypting it using cryptography
methods. Early steganographic algorithms considered
only human abilities to spot Irregularities as the only
detection technique. Those algorithms implementations
are widely used in Image Steganography and relies on
the fact that computer images normally have quite a bit
of redundant data and that changing the contents of
those data (as pixels or color plate elements) could make
us enough space to embed a considerably large message
(50% data rate with BPCS Steganography [5]), where
BPCS stands for Bit-Plane Complexity Segmentation
[5]. Not all early Steganographic algorithms had huge
data rate of the BPCS, but most of all algorithms
perform their data embedding in very unique manner
that create some unique irregularities to distinguish each
algorithm from others.
Classification of Image Steganographic techniques
can be done based on which domain has to be used
during embedding the secure message into two groups:
Spatial/image Domain and Frequency/Transform
Domain. Spatial domain techniques embed the secure
message stream in the image pixels directly, while in
frequency domain, images are first transformed from
spatial to the frequency domain and then the message
steam is embedded in the transformed form of the image
[2].
A. Spatial Domain steganography
First generation of steganography, embedding
process uses the spatial domain of the image to embeds
the message data in a sequential order in the Least
Significant Bit of image pixels [16][17]. One of the
disadvantages of this technique is the weak immunity
against compression where there is a fear for damage of
the message that may have sensitive information [11], in
addition to the poor immunity against visual attaches
and the simplicity to extract the embedded message by
walkthrough around image pixels. To decrease the
ability on extracting the embedded message, A Pseudo-
Random Number Generator (PRNG) works based on
input parameter used as a secret key in order to generate
randomized ordered locations to be used for hiding the
message bits instead of the sequential order of message
embedding [18].
Moreover, in 2003 Rios and Puech[13] introduce a
new approach depending on imperceptible of human eye
to a small variation in the channel color value of a
colored image [14] and this approach called SSB-4
steganography approach [15]. The main concept in SSB-
4 is to embed the secure message into the 4th bit of the
original image pixel, and to minimize the value
difference between the modified pixel value that have
the secret message bit and the original pixel value,
modify the 1st, 2nd, 3rd and/or 5th pixel values to align
modified pixel value to the original value.
Unfortunately, LSB didn't stand forever as new
evolving branch of data analysis was born
"Steganalysis". The first released results of steganalysis
was a paper by Andreas Westfeld and Andreas
Gtzmann[6]. Gtzmann and Westfeld made clear how to
attack a number of very famous image based
steganography algorithms both visually by the use of
some image techniques and naked eye, or by automated
statistical algorithms. The two attacks introduced were
the filter attack and the PoV statistical attack [6]. Both
attacks were designed to address the Steganographic
systems of spatial domain embedding.
B. Frequency Domain Steganography
Unlike spatial domain, the data embed into image
pixels directly, in transform domain, image
representation is transformed to frequency domain
before start hiding message data into image [6][10]. The
main factors to determine image payload to embed the
message are the number of pixels and the color depth
with taking into consideration preserving the invisibility
of image data changes during embedding process [3].
it’sworthtomentiontheobservationofWang’sinthere
paper [21] where they find data embedding in the
frequency domain, cause the hidden data spread across
the entire image in more robust areas, and gives a better
immunity against signal processing[21].
Discrete cosine transform (DCT) is the most well-
known transform coding techniques that is widely used
in converting a signal into elementary frequency
components used to implement lossy image
compression (such as JPEG format) to transform the
image from the spatial domain to the frequency domain.
DCT separates the image into three different frequency
components: high, medium and low where the image is
segmented into non-overlap 8 pixels x 8 pixels blocks.
The DCT is computed for each block starting from left
corner to right corner in top-bottom order [13]. In this
paper, we consider the DCT as an example of the
frequency transformation technique that can be used.
To enhance the security of embedding using LSB
a Pseudo-Random Number Generator (PRNG) and a
secret key have been used to specify the order of access
to the embedded information as introduced by F5
Algorithm according [17] where a user password in
included algorithm as input to be used for PRNG. The
3. embedding process uses the user password to get the
seed for PRNG and then to generating a random walk
through the DCT coefficients of the cover image
Another approach uses the randomization in
embedding the secret message introduced by Bani
Younes et al[4] who used the encrypted image to send
the secret information through it. The basic of this
method is to mixed the generate number of horizontal
and vertical blocks at the sender side with the encrypted
image before sending it to the intended party. In this
method, the authors select randomly based on a user
input key a number of bytes to replace there LSB with
the binary representation of the hidden data within the
encrypted image and by this way the hidden information
will spread among the encrypted image data randomly
based on the generated locations by the user secret key.
Following the attacks by Gtzmann and Westfeld a
new Steganographic system emerged to use the
transformed image to frequency domain using discrete
cosine transformation algorithm as carrier for embedded
data instead of embedding the message directly to the
pixels in spatial domain [6][10].Steganographic methods
that use embedding in frequency domain have the
immunity against PoV and filter attack, but had a lower
data rate. OutGuess and F5 can be considered as an
example for this implementation [7, 8] where F5 use the
DCT coefficients to embed to embed the secure message
inside them with excluding some coefficients values
such as zero AC coefficients. The concept of F5
algorithm is instead of replacing the LSBs of quantized
DCT coefficients with the message bits, the absolute
value of the randomly selected coefficient is decreased
by one. F5 authors claimed that this type of embedding
cannot be detected using their ᵡ2 statistical attack [18].
In order to increase complexity against steganalysis,
instead LSB, hidden data can be embedded in the Bit-4
of the DCT coefficient [10].
C. Stego Image Quality Measure
The idea behind steganography is to preserve the
secrecy of embedded message by hiding every existence
of the concealed message. As the obscurity is the art of
Steganography, the human perception immunity of the
stego-image against detection is the most important test.
The most used tests are the Subjective in addition to
the mathematical tests, Peak signal-to-noise ratio (PSNR
calculated in dB (decibel) ) is a commonly
used measurement of the picture degradation, which is
calculated as the error between the original image and
stego-image[11].
ITU (International Telecommunication Union)
defined a set of rules and recommendation of subjective
tests. The subjective tests are carried out by people who
look for visual differences between the original and the
stego- image to trying to. If the percentage of success
greater than 50%, it can be concluded that the message
is invisible [14][2].
Peak Signal to Noise Ratio (PSNR) is a technical
approach usually used to evaluate the real quality of
stego image [13][4]. This technique is an engineering
term for the ratio between the maximum possible power
of a signal and the power of corrupting noise that affects
the fidelity of its representation. The PSNR is most
commonly used to measure the quality of reconstruction
in an image; by comparing the stego image with the
original image.
III. THE PROPOSED METHOD
Our contribution will be introducing a Pattern
Matching algorithm to be used during Steganography
embedding process, this will enable finding reasonable
match between the cover image and the message binary
stream which will result in minimizing distortion and
modification that can be caused by modifying the image
data to hide the secrete message data by Steganography
algorithm. We are proposing an improved
implementation for steganography algorithm on top of
DCT (Discrete Cosine Transformation) frequency
domain where the implementation will includes
implementation for Pattern Matching algorithm, where
this algorithm will try to generate random message
steam data locations based on random keys generator
that keep working until reach certain threshold value or
maximum number of trials, below is the pseudo-code for
the proposed algorithm. Thus will increase the ratio for
of the data to be embedded with taking advantages from
the face of extremely low embedding rate prevents all
known statistical attacks[9].
In Figure 1, the general description to depict the
process of the proposed method. The first step is to
partitioning the image into non-overlapped 8x8 blocks,
to transfer the image blocks to frequency domain, we
apply DCT on each block. When the corresponding
DCT blocks computed for the original image blocks, the
pattern matching algorithm tries to find the most
reasonable match between DCT LSB values and the
message bits that we want to hide. Once the
reasonable match has been found, message data
embedded into least significant bit of the nonzero
DCT coefficients. Lastly we apply the IDCT on each
block to producing the stego-image which can be
transferred to the intended recipient. Our approach is
illustrated in details in the following five steps
(algorithm):
4. Figure 1. The process of the proposed method
Step 1: Applying 2D DCT On Image Pixels
In this step, the image is partitioned into non-
overlapped 8x8 blocks. Thus each block F(x,y) consists
of 64 pixels values. In case the image is 8 bits depth
monochromic, then F(x,y) consists of the whole pixels'
values. And if the image is RGB 24-bits depth colored,
then each block F(x,y) is constructed from only the least
significant bytes (i.e., Blue color channel/contribution)
of the successive pixels. This is because the Blue
channel is the most imperceptible to human eye. Then
we calculate the 64 DCT coefficients S(v,u) of each grid
F(x,y) of the image using DCT equation[20].
Step 2: Finding The Reasonable Pattern Match
Between DCT Blocks And Message Data
Herein the image is transferred into frequency
domain/DCT blocks, the pattern matching algorithm is
started to enable finding accepted matching between the
cover image and the message binary stream which will
result in minimizing distortion and modification that can
be caused by modifying the image data to hide the
secrete message data. Our pattern matching algorithm is
illustrated in details in the following code:
1.ReasonablePatternMatch() : INTEGER /*Return the
number that generate the reasonable match
between the random locations in the image
and the message*/
2.BEGIN
3.INTEGER BFK= 0; /*choose initializing seed to
generate rando locations*/
4.INTEGER bestGainedKey = BFK;/*supposethatit’s
the best generated key*/
5.INTEGER bestError = 99999999; /*Initial value
where this vaiable will be update if the tested generated
key produce less location mismatch between the
generated locations in the stegoimage and the
message*/
6.WHILE TRUE LOOP
6.1.INCREMENT(BFK);
6.2.order = getOrder(BFK);/*Get random
locations based on BFK value*/
6.3.FOR each bit in msg data LOOP /*loop to
compare the msg data and image data */
6.3.1.IF msg bit not equal LSB LSB in
image[order[i]] THEN /*(image[order[i]]>>1<<1 |
message[i] != image[order[i]]) compare the message
bit with the most left image bit, if they are not similier
then this will count as error since when we embed the
message in the image then such image data will be
chaged according ti the message data*/
6.3.1.1. INCREMENT(error); /*mismatch
counter increment*/
6.3.2.END IF
6.3.3.IF error less than bestError THEN/*if we get
minimum error rate compared with prev. error rate*/
6.3.3.1.bestGainedKey = BFK; /*update the best
generated key*/
6.3.3.2.bestError = error; /*update the
minimum error rate we get*/
6.3.4.END IF
6.4.END LOOP
6.5.IF bestError less than MaxError THEN /* if we
reach the acceptable error rate*/
6.5.1Return bestGainedKey; /* return the key to
be used*/
6.6.END IF
7.END WHILE
8.RETURN bestGainedKey; /* return the key to
be used*/
9.END
Reasonable Pattern Matching Algorithm(RPMA)
In the above Reasonable Pattern Matching
Algorithm (PRM) uses a supplementary algorithm
called in step 6.2 named getOrder to return a list of
random locations that will be used to find the locations
in the stego-image that will be used to hide the message
data inside it.
Step 3: Embedding Message Bits
At this stage, to start message bits embedding, the
generated random locations found in step 2 will be used
to embed the message into the DCT coefficients based
on the generated sequence. Message will be embedded
into the least significant bit of the DCT coefficients with
exclude the zero DCT values.
Step 4: Construct The Stego Image
Lastly, the stego-image is constructed by replacing
each original image block F(x,y) by the modified stego
block F'(x,y) where the stego-image contains the image
data in addition to the hidden message data.
5. Unlike the other techniques such as F5, our
approach will have no significant effect on the DCT
blocks of the stego-image F'(x,y) generated in step 3,
because in step 2, the pattern matching algorithm will
try to find the appropriate locations inside the cover
image to hide the secrete message taking into
consideration the minimum number of stego-image
DCT coefficient changes.
IV. EXPERIMENTAL RESULTS
Our contribution will be introducing Pattern
Matching algorithm to be used during steganography
embedding process, this will enable finding a better
matching between the cover image and the message
binary stream which will result in minimizing
distortion and modification that can be caused by
modifying the image data to hide the secrete message
data. To test the actual validity of our Pattern Matching
algorithm, we implemented an algorithm to measure the
probability to find a pseudo random locations in the
cover image we can hide the secrete message in them
with minimum changing in original image data.
To validate our algorithm and results, we have
selected four different images with same dimension size
to be used as the container that the secure message will
be embedded inside it. And also we have selected two
secrete text messages with different sizes 10 and 20
bytes respectively, to test our approach we deal with the
text message as a binary steam and start our experiment
by first image by reading the Original Image, transfer
it to DCT image data, generate random locations in the
DCT image data based on random number generator to
be used as a location for hiding the secrete message,
after that find the middle frequency values[22] from
these locations, compare them (the least significant bit
value of these DCT data)with the binary stream of
secure message data and recording the match and
mismatch between both image and message data. We
have tested the similarity ratio improvement between the
message and the random selected image data by making
the algorithm to be run on five stages: first stage run the
algorithm for 50 times to find the similarity and then run
it again up to 200 times, 400, 800 and 1600 time to
measure the effectiveness of finding best matches
between the secrete message data and image,
following finger2and3show our experiments results
for 10 and 20 bytes message size respectively.
Experimental statistics on the proposed pattern matching algorithm
10 bytes message
Image Name Dimension
Pattern matching
algorithm Iteration
Worst generated key
Similarity ratio
Worst key
Best generated
key
Similarit
y ratio
Best key
Flowers.jpeg 400 x 300 50 1932935652 21% -946158187 41%
Flowers.jpeg 400 x 300 200 305530653 18% -1816578671 43%
Flowers.jpeg 400 x 300 400 1917256675 16% -1494646998 45%
Flowers.jpeg 400 x 300 800 -1584762429 13% 1708490964 46%
Flowers.jpeg 400 x 300 1600 -1471890995 10% -488980360 47%
People (Lena) .jpeg 400 x 300 50 -1582118205 20% 1852959459 43.75%
People(Lena) .jpeg 400 x 300 200 -1582118205 20% 1852959459 43.75%
People(Lena) .jpeg 400 x 300 400 -1087887718 16.25% 1852959459 43.75%
People(Lena) .jpeg 400 x 300 800 -1087887718 16.25% -1087887718 46.25%
People(Lena) .jpeg 400 x 300 1600 2117760478 16.25% -1087887718 46.25%
Bridge(Landon).jpeg 400 x 300 50 -1912515890 36% -1111958610 53%
Bridge(Landon).jpeg 400 x 300 200 306857764 35% -1111958610 53%
Bridge(Landon).jpeg 400 x 300 400 306857764 35% -1111958610 53%
Bridge(Landon).jpeg 400 x 300 800 306857764 35% -1111958610 53%
Bridge(Landon).jpeg 400 x 300 1600 306857764 35% -1111958610 53%
Mountains.jpeg 400 x 300 50 -605538897 20% -933251167 41%
Mountains.jpeg 400 x 300 200 -605538897 20% 125526107 45%
Mountains.jpeg 400 x 300 400 965414341 18% 125526107 45%
Mountains.jpeg 400 x 300 800 965414341 18% -883571161 46%
6. Mountains.jpeg 400 x 300 1600 445598606 13% -883571161 46%
Figure 2. Experiments results for 10 bytes message size
20 bytes message
Image Name Dimension
Pattern matching
algorithm Iteration
Worst generated
key
Similarity ratio
Worst key
Best generated
key
Similarit
y ratio
Flowers.jpeg 400 x 300 50 2002860108 24% 1644278906 42%
Flowers.jpeg 400 x 300 200 -655012674 23% 1644278906 47%
Flowers.jpeg 400 x 300 400 281662087 21% 1644278906 44%
Flowers.jpeg 400 x 300 800 281662087 21% 1644278906 47%
Flowers.jpeg 400 x 300 1600 1368526818 19% 1644278906 47%
People (Lena) .jpeg 400 x 300 50 387420073 20% -648114300 40%
People(Lena) .jpeg 400 x 300 200 387420073 20% -648114300 40%
People(Lena) .jpeg 400 x 300 400 387420073 20% -648114300 40%
People(Lena) .jpeg 400 x 300 800 387420073 20% 1079306710 41%
People(Lena) .jpeg 400 x 300 1600 387420073 20% 1879938708 49%
Bridge(Landon).jpeg 400 x 300 50 397203127 43% 1891607262 53%
Bridge(Landon).jpeg 400 x 300 200 1963787566 42% -69194182 54%
Bridge(Landon).jpeg 400 x 300 400 1963787566 42% -69194182 54%
Bridge(Landon).jpeg 400 x 300 800 -1425495006 41% -69194182 54%
Bridge(Landon).jpeg 400 x 300 1600 -1425495006 41% 1333860696 56%
Mountains.jpeg 400 x 300 50 -1437932735 23% 926323778 38%
Mountains.jpeg 400 x 300 200 1483788745 19% -736319659 42%
Mountains.jpeg 400 x 300 400 1483788745 19% -736319659 42%
Mountains.jpeg 400 x 300 800 1483788745 19% -736319659 42%
Mountains.jpeg 400 x 300 1600 1483788745 19% 1087559179 43%
Figure 3. Experiments results for 20 bytes message size
According to our observation and stated results in tables above, we can say that using pattern matching
algorithm will be reasonable and fructiferous idea as this algorithm will be used to reduce the needed number of
changes to hide a message inside cover image taking into account the length of the message where the length of
the message to be hidden are In direct proportion with computations needed to find best matching positions in
the cover image. Below is the sample images we have used in experiments:
Figure 4. Images that have been used in previous experiments(experiments results in figure 2 and 3)
As a visual example, the bellow figure 5 A is the source image that used by our Steganographic algorithm to
produce finger B the stego-image. It is clear that the stego-image are imperceptibleand also has a very good PSNR
which is equal to 54.6508.
7. Figure 5.A Original Image Figure 5.B Stego-image
V. CONCLUSION
In this paper we present an enhanced algorithm for
Steganography, this algorithm that we claim to be safe,
built over DCT (Discrete Cosine Transformation)
frequency domain mutation, the algorithm uses pattern
matching approach to obtain a reasonable a better image
quality by reducing number of changes that
steganography algorithm made during the embedding
process in carrier which will lead to finding less
differences between the stego-image and original one.
According to our observation and obtained results,
pattern matching algorithm will be a reasonable and
fructiferous idea as the algorithm will reduce the needed
number of changes to hide a message inside cover
image which is the very important to reduce image
distortion and preserve a good image quality compared
to the original cover image.
REFERENCES
[1] Wikipedia contributors. Steganography. Wikipedia,
the Free Encyclopedia; 2011 .
[2] J. Lenti, “Steganographic methods,” periodic
polytechnic ser.el.eng 44, No.3-4 , pp 249-258, Jun
2002.
[3] C. C. Chang, T. S. Chen, and L. Z. Chung, “A
steganographic method based upon JPEG
and quantization table modification” Information
Sciences, Volume 141, Issues 1-2, Pages 123-
138, March 2002.
[4] M. A. Bani Younes, A. Jantan, “A New
Steganography Approach for Image Encryption
Exchange by Using the Least Significant Bit
Insertion”, IJCSNS, International Journal of
Computer Science and Network Security, Vol. 8
No. 6, June 2008.
[5] Eiji Kawaguchi and RichardO.Eason,“Principle
and applications of BPCS-Steganography”,
University of Maine and Kyushu Institute of
Technology, 2000
[6] Niels Provos, “OutGuess-Universal
Steganography”, outguess.org, August 2001
[7] A. Westfeld, “F5 Steganographic Algorithm: High
Capacity Despite Better Steganalysis, Proc”. 4th
Intil Work shop Information Hiding, Springer-
Verlag, pp 289-302, 2001.
[8] N. Provos and P. Honeyman, “Detecting
Steganographic Content on the Internet”, Proc.
2002 Network and Distributed System Security
Symp.,Internet Soc, 2002.
[9] J. Fridrich, M. Goljan, R. Du, "Reliable Detection
of LSB Steganography in Color and Grayscale
Images", workshop on Multimedia and security:
new challenges, Pages 27 – 30,2001
[10] Kafri N. and Sulaiman H., “Bit-4 of Frequency
Domain-DCT Steganography Technique”,
Networked Digital technology NDT2009, VSB
Technical University, Czech Republic, 29-31, July
2009.
[11] N.Ahmed,T.Natarajan,andK.R.Rao,“Discrete
cosinetransform,”IEEE nuns. Compur., vol. C-23,
pp. 90-94, Jan. 1974.
[12] Abbas Cheddad et al,” Digital image
steganography: Survey and analysis of current
methods, signal Processing 90 ,727–752,2010
[13] J. Rodrigues, J. Rios, and W. Puech “SSB-4
System of Steganography using bit 4”, In
International Workshop on Image Analysis for
Multimedia WIAMIS, (Montreux),May 2005.
[14] International Telecommunication Union,
“Information Technology- Digital Compression
and Coding of Continuous-Tone Still Images -
Requirements and Specifications Recommendation
T.81”,ITU, Sept 1992
8. [15] IsmailAvcibasN.M.andB.Sankur,“Steganalysis
using image quality metrics”, In IEEE
Transactions on Image Processing, vol. 12, No. 2,
February 2003.
[16] L. Davidson, and P. Goutam, “Locating secret
message in images”, In ACM SIGKDD
international conference on Knowledge discovery
and data mining, (Seattle, Washington, Aug. 22-
25). ACM 1-58113-888-1, 2004.
[17] J. Fridrich, M. Goljan, “Steganalysis of JPEG
Images: Breaking the F5 Algorithm”, Publisher:
Springer Berlin, Heidelberg, Lecture Notes in
Computer Science, pp 310-323, Vol. 2578/2003.
[18] Anjali A. Shejul, Umesh L. Kulkarni, “A Secure
Skin Tone based Steganography Using Wavelet
Transform”. International Journal of Computer
Theory and Engineering, Vol.3, No.1,PP 1793-
8201, February 2011
[19] Tiwari1 N., Shandilya M., “Secure RGB Image
Steganography from Pixel Indicator to Triple
Algorithm-An Incremental Growth”, International
Journal of Security and Its Applications Vol. 4, No.
4, October 2010
[20] Neil R. Bennett, “JPEG STEGANALYSIS &
TCP/IP STEGANOGRAPHY”, UNIVERSITY OF
RHODE ISLAND, 2009.
[21] G. Langelaar, I. Setyawan, R.L. Lagendijk,
“Watermarking Digital Image and Video Data”,
IEEE Pattern Processing Magazine, Vol 17, pp 20-
43. ,2000
[22] S. Dickman, ”An Overview of Steganography”,
Research Report, James MadisonUniversity 2007,
JMU-INFOSEC-TR-2007-002, July 2007.