Stegnography of high embedding efficiency by using an extended matrix encoding algorithm

IJRET: International Journal of Research in Engineering and Technology ISSN: 2319-1163
__________________________________________________________________________________________
Volume: 02 Issue: 05 | May-2013, Available @ http://www.ijret.org 858
STEGNOGRAPHY OF HIGH EMBEDDING EFFICIENCY BY USING AN
EXTENDED MATRIX ENCODING ALGORITHM
Borkar Bharat Sampatrao1
, Patil Pritesh Kashinath2
1
Assistant Professor, 2
Student, Department of IT, AVCOE, Sangamner, Maharashtra, India
borkar.bharat@gmail.com, mr.pkpatil1989@gmail.com
Abstract
F5 Steganography is way totally different from most of LSB replacement or matching steganographic schemes, as a result of matrix
encryption is used to extend embedding potency while reducing the amount of necessary changes. By victimisation this theme, the
hidden message inserted into carrier media observably is transferred via a safer imperceptible channel. The embedding domain is that
the quantitative DCT coefficients of JPEG image, which makes the theme, be proof against visual attack and statistical attack from the
steganalyst. Based on this effective theme, An extended matrix encoding algorithm is planned to improve the performance further in
this paper. The embedding potency and embedding rate get accrued to large extent by changing the hash function in matrix encryption
and changing the coding mode. Eventually, the experimental results demonstrate the extended algorithm is more advanced and
efficient to the classic F5 Steganography.
Index Terms: Steganography, LSB replacement, DCT coefficient, Hash function.
-----------------------------------------------------------------------***-----------------------------------------------------------------------
1. INTRODUCTION
Steganography is the art and science of writing the secret
content inside cover media and transferring the stego media
from the sender to intended recipient through a subliminal
channel without arousing the suspicion of adversary. The
presence of hidden info is meant to be undetectable. If the
actual fact that communication is happening is revealed, the
steganography is cracked not withstanding whether or not or
not the hidden info is exposed. Thus, compared with other
connected techniques like watermarking, the property of
covertness plays a crucial role within the stegosystem.
In order to create stegotext apparently innocent, the
confidential message is typically embedded into the redundant
components of cover media. For digital image, the least
significant bit plane in spacial domain is one reasonably these
components that appear as if completely random and noisy.
The modification of LSB won't cause noticeable change of the
looks of image. Several LSB based techniques of data hiding
are proposed in recent years [2,3]. Derek Upham’sJSteg was
most likely the primary in public accessible steganographic
system for JPEG images [4]. This technique is actually a copy
of the LSB substitution algorithm in spacial domain. The
least-significant bit of DCT coefficients is consecutive
replaced with the secret message. Since the replacement solely
happens on 2 adjacent coefficients, it'll cause a statistically
obvious POVs (pairs of values) problem which may be with
success detected by X2
-test proposed by Westfeld and P
fitzmann [5].
2. BACKGROUND
To improve the encoding step by scattering the embedding
locations over the complete DCT-domain in keeping with a
pseudo-random number generator, the Outguess0.1 is
developed [6]. The randomly distributed data cannot detected
by the X2
-test for JSteg successfully. The creator of
Outguess0.1 releases a revised version soon. The new
algorithm known as Outguess0.2 tries to make sure that the
statistics properties of the cover image may be maintained
after encoding [7]. Shortly, Fridrich et al. use the discontinuity
of the border between 2 adjacent eight by eight blocks to
accurately estimate the length of the hidden information
embedded with Outguess0.2 [8]. As another to the
Outguess0.2 algorithm, Westfeld invent the more secure
algorithm called F3 algorithm[9]. This method compares the
absolute value of the coefficients with the secret message. If
the absolute value is same, then no modification is done.
Otherwise, The corresponding DCT coefficient’s absolute
value is reduced by one. F3 steganography eliminates the
POVs problem which exists in JSteg by using above
embedding strategy. In F3 the distribution of histogram looks
unnatural because in F3 there are more even coefficients are
introduced into histrogram than odd coefficients. Hence, to
deal with unstructured frequency distribution of histogram an
improvement called F4 is developed. In F4, even-negative
coefficients and odd-positive coefficients represent one and
odd-negative coefficients and even-positive coefficients stands
for zero. If the secret bit value is same as the symbolic
coefficient, then no changes happen. Otherwise, absolute
value of the corresponding DCT coefficient is reduced by one.
With the assistance of this embedding strategy, the histogram

__________________________________________________________________________________________
of stego image looks statistically like a clean image.
Additionally, there are still several steganographic algorithms
proposed like model-based [10,11], Perturbed quantization
[12], YASS [13] so on [14–18].
Nonetheless, Many steganographic algorithms consider the
embedding capacity but but neglect the embedding efficiency
and security. A number of them simply work oppositely.
During this paper, by improving the matrix encoding of F5
stegosystem to increase the stegosystem we propose an
extended algorithm, while the embedding rate is essentially
improved likewise. Some changes in hash function in matrix
encoding are performed to cover additional secret bits into
definite range of cover bits. By exploiting the n-layer
extension, we convert the triple (dmax , n, k) to quad
(dmax , n, k, L) to extend the embedding efficiency and
embedding rate at same time. To indicate current number of
layer, the symbol positions are deployed in every embedding
process. This mechanism makes the receiver understand blind
detection effectively. the rest of this paper is organized as
follows. In Section two, the quality matrix encoding is
represented. In the next section we represent the details of the
proposed extended algorithm. In Section four, some
simulation results and analysis are given, and final conclusion
is provided in Section five.
3. EXISTING SYSTEM
3.1 Matrix Encoding Basics
The F5 scheme is developed with the help of F3 and F4 which
are developed by JSteg, The F5 scheme is developed by
Westfeld [9]. We insert the secret message by modifying
corresponding LSB positions of quantized DCT coefficients of
image, JPEG image as well. Instead of classic LSB
replacement or matching methods, the matrix encoding in F5
are used to implement the insertion and detection of secrete
message. To enhance the security of stegosystem, the
additional ‘permutative straddling’ technique is used to scatter
the secret message over the whole carrier media.
Matrix encoding, as a lower rate method improves the
embedding efficiency to a great extent than the classic LSB
modification methods. It is introduced by Crandall [19]. The
embedding efficiency depends on, how many bits of secret
message can be loaded by one change taken place in carrier
media. Reference from the LSB method, assume that the
secret message and the values at the positions to be changed
subject to the 0–1 uniform distribution. The positions which
are to be embedded modified with a probability of 0.5 on an
average. Means we have an embedding efficiency of 2 bits per
change. If we want to fully embed the whole cover image,
50% of the carrier data will be changed. And it is so usefull in
a high possibility of statistical stegnalysis. By reducing the
density of changes in the cover image we reduce the
possibility of detection, this is the most obvious and well
known way as ever. Matrix encoding decreases the necessary
number of changes (i.e the change density), as well as it
increases the embedding efficiency. The description of the
encoding process is given below:
1) Implement the JPEG lossy compression on the
carrier image and obtain the quantized DCT
coefficients for embedding [20].
2) The LSB plane of quantized DCT coefficients is
partitioned into many embedding cells which are
in the form of vector 𝑎 = 𝑎1 𝑎2. . . 𝑎 𝑛 with the
length of n.
The coding which implemented on each embedding cell is
denoted by an ordered triple (dmax , n, k) . where, n is the
number of modifiable bit positions in an embedding cell,
namely, the length n. k is the bit length of secret message w =
w1w2. . . wk to be embedded into one cell. Also, an
embedding cell with n positions will be changed in not more
than dmax positions to embed k bits secret message. F5
implements matrix encoding only for dmax = 1 . A hash
function f is defined as Formula (1) to map n bits cover data a
into k bits binary string.
f: f a =
n
⊕
i = 1
ai. i, (1)
where ai denotes the i-th position of cell a. i is the
corresponding index number of the bit position and is in
binary form during the operation. The bit length of binary i is
selected as the same size as secret message w. Subsequently,
implement XOR operation on the value of hash function and
secret message w to obtain a decimal number y.
y = w ⊕ f(a) (2)
By logically flipping the y-th bit position of cell a , an
embedded stego data a′ is generated. In special case, the
carrier cell a′ will be left intact wheny = 0.
a′
=
a if y = 0
a1a1 … ay … an otherwise
(3)
Where, ay is the negation ofay.
In extraction phase, the receiver would retrieve the secret
message w by directly putting the stego cell a′
into the same
hash function f. The detailed procedure of matrix encoding is
given in Fig. 1.
For an embedding cell, the change density D refers to the
proportion of altered bit position. Neglecting shrinkage, we
can calculate the change density depending on Formula (4):
D =
n
n+1
.1+
1
n+1
.0
n
=
1
n+1
(4)

__________________________________________________________________________________________
We can also get another important performance factor of
steganographic algorithm called the embedding rate R:
R =
k
n
(5)
Using the change density and embedding rate, we can
calculate the embedding efficiency E which indicates the
average bit number of embedded secret message per change:
E =
R
D
=
n+1
n
. k (6)
Fig -1: The block diagram of matrix encoding.
The theoretic values of the change density are given in Table
1, the embedding rate and embedding efficiency in according
to various of (n,k) pairs. The values are calculated by using
Formulas (4)–(6).
In F5 algorithm, the value of n and k must satisfy the equation
n = 2k
− 1. Because k-bit binary sequence has 2k
possible
states in all. It is inevitable that the binary carrier cell has the
length of 2k
− 1 to show all states of k bits secret message by
means of modifying only 1 bit or keeping unchanged. That is
to say, the length of embedding cell increases with the
increase of the length of embedded secret message in manner
of exponential growth. Meanwhile, the modification of carrier
cell maintains 1-bit-change. Thus, the embedding rate will
getting less and less with increase of efficiency. The value of
higher efficiency is the lower rate. This regulation not only
reveals how matrix encoding works, but also naturally inspires
us that we are able to improve both the embedding efficiency
and the embedding rate at the same time if k increases in the
context of fixing on n and the number of change.
4. PROPOSED SYSTEM
4.1 Extended Encoding Algorithm
How to do the independent increase of k? In matrix encoding,
the hash function maps n bits carrier data into a certain length
of binary sequence that depends upon the bit length of index i.
And the bit length of i is chosen as the same size as that of
secret message w. Given the extension of the length of i is
realizable, we will finally embed additional bits of secret
message into one cell. Taking the cell (1,3,2) as associate
example, 2-bit secret message will be embedded into 3-
bitembedding cell by changing only 1-bit position of the cell.
The length of i is two. If the length of i is extended to be 3-bit,
we can take one additional bit of secret message to implement
exclusive-or with the binary result of hash function. However,
the problem shows up. The result of XOR operation, namely,
the index y which will be used to indicate the position to be
modified is out of the range. We may get y = (101)2 = 5 in
that case, however it’s not possible to find out the fifth bit
position for a cell of 3-bit length. Essentially modifying just 1
bit or keeping unchanged in the carrier cell with 3-bit length
can only express four kinds of secret code. These type of code
are named as ‘00’, ‘01’, ‘10’, ‘11’ with the length of log2 4.
The secret code extended to 3 bits has 23
states in all. Thus,
there is no way to embed all of eight kinds of code into 3-bit
cell by using matrix encoding algorithm.
Actually, there is indeed a way to extend but require to select
some extended codes elaborately. The extension appears to be
conditional. Since 3-bit modifiable cell is only able to express
four states, we still have a half opportunity to extend by
selecting four extended codes to embed from eight codes.
During calculating the result of hash function, we can simply
multiply the index i by 2 to extend 1 bit where we call it 1-
layer extension. In this case, the codes ‘00’, ‘01’, ‘10’, ‘11’
are extended to ‘000’, ‘010’, ‘100’, ‘110’. In a similar way, we
can multiply i by 22
to extend 2 bits called 2-layer extension.
The rest may be deduced by analogy. L-Layer extension is
performed by multiplying iby 2L
. Due to the closure property
of XOR operation,
Table -1:The performance of matrix encoding.
(𝐧, 𝐤) 𝐃(%) 𝐑(%) 𝐄(bit)
(1,1) 50 100 2
(3,2) 25 66.67 2.67
(7,3) 12.5 42.86 3.43

__________________________________________________________________________________________
(15,4) 6.25 26.67 4.27
(31,5) 3.13 16.13 5.16
(63,6) 1.56 9.52 6.09
(127,7) 0.78 5.51 7.06
(255,8) 0.39 3.14 8.03
(511,9) 0.2 1.76 9.02
we can embed the secret message with more than 2-bit length
into 3-bit cell, provided that the secret code is equal to any one
of the specific extended codes. The mode of extension is
illustrated in Fig. 2.
For extended algorithm, the coding mode implemented on the
embedding cell is redefined by a quad (dmax , n, k, L) ,
wherethe new parameter L denotes the maximum of extension
layer. Firstly, take out (k + L) -bit secret code w =
w1w2 . . . wk … wk+Lfrom the whole secret message sequence
to test if the secret code matches a specific extended code in
the L -th layer. Thematching method is to test whether
mod w, 2L
= 0 is true. If the remainder equals to zero, the
extension layer of currentcell lcrt is L and a (k + L)-bit secret
data will be able to be embedded successfully. If not, then
continue to test if the prior (k + L − 1) -bit secret code
w = w1w2 . . . wk … wk+L−1 matches a specific extended
code in the (L − 1) -th layer by testing the resultof
mod w, 2L−1
. If mod w, 2L−1
= 0 is true, then the current
extension layer is lcrt = L − 1 and the secret code w =
w1w2 . . . wk … wk+L−1 will be embedded into this cell. But if
not, continue to do this kind of test until we find out a
matching code in a certain layer or there is no matching code
in all extension layers. In latter case, the extended algorithm
rolls back to the standard matrix encoding. The final
embeddable secret code is in the form of
w = w1w2 . . . wk … wk+lcrt
. If no extension takes place, the
layer of current cell is lcrt = 0.
In extended algorithm, the hash function is updated as
Formula (7):
f: f a =
n
⊕
i = 1
ai. (i. 2lcrt ) (7)
Subsequently, implement XOR operation on the result of hash
function and secret message w = w1w2 . . . wk … wk+lcrt
to
obtain a decimal number y. At the moment, the range of index
y has already been extended. We must shrink it to n by
makingy divided by a coefficient 2lcrt .
y =
w⨁f(a)
2lcrt
, (8)
where the result of w⨁f(a) is expressed as a decimal number.
Eventually, we obtain a stego cell a′ by negating the y-th
position in carrier cell a.
a′
=
a
a1a1 … ay … an
ify =0
otherwise
(9)
From the above statement, it is implied that the introduction of
extension mechanism raises a new problem to the receiver in
detection process. The coding quad (dmax , n, k, L) can be
confirmed and shared by the sender and the receiver before the
start of the communication. Since the current layer lcrt is
relative to the content of secret message, the receiver cannot
predict this parameter definitely. Accordingly, the sender has
to transfer lcrt to the receiver in the embedding process. We
decide to append a symbol s = s1s2. . . sm to the stego cell
a′
= a1a2 … ay … an to mark the layer lcrt . Because the value
of lcrt is fallen into the closed interval of [0,L], the length of
symbol m can be calculated by Formula (10). We use the
binary number of lcrt to assign the symbol s.
m = [log2(L + 1)] (10)
Thus, the new stego cell c with the length of (n + m) is
composed of two parts, namely, data part and symbol part (i.e.
thecell is reformed as c = a′
s = a1a2 … ay … ans1s2 … sm ). In
extraction phase, the receiver firstly take out the symbol part
of thestego cell c and calculate the layerlcrt .
lcrt = dec(s1s2. . . sm ) (11)
Fig -2: The chart of the extension mode.
Eventually, the extended secret data
w = w1w2 . . . wk … wk+lcrt
is retrieved by putting the data
part of the stego cell c intothe updated hash function f: f a =
n
⊕
i = 1
ai. (i. 2lcrt )
w = f(a′) (12)

__________________________________________________________________________________________
The detailed procedure of the extended matrix encoding is
shown in Fig. 3.
To be more clear, we take the coding mode of (1,7,3,2) as an
example to show how the extended algorithm works. Assume
that the carrier data is a = 1101010, the secret data taken
from the whole secret sequence is w = 11001.
First of all, the sender tests if mod dec 11001 , 22
= 0 is
true. Due to mod dec 11001 , 22
= 1, the sender continue
to testthe shorter secret data w = 1100 . Since
mod dec 1100 , 21
= 0 is true, it is confirmed that the
secret data w = 1100 can be embedded into the carrier data
and the current layer lcrt = 1.
Secondly, calculate the length of symbol m = log23 = 2
and assigns the symbol s = 01.
Thirdly, calculate the hash function with the carrier data
a = 1101010 as shown as follows:
Fig -3: The flowchart of the extended matrix encoding.
0011
0100
1000 < ⋯ ai. (i. 21
)
f a :⊕ 1100
______
0011
(13)
Finally, calculate index
y = (w ⊕ f(a)) 2 = ((1100)2 ⊕ (0010)2) 2 = 7 and flip
the seventh bit position of carrier data a = 1101010 to
generate a stego data a′ = 1101011. Up to now, a stego cell
c = 110101101 is obtained.
In detection process, the receiver firstly takes the symbol
s = 01 from the stego cell and calculates the current layer
lcrt = dec(01) = 1 . Subsequently, calculate hash function
with the stego data a′ = 1101011 to retrieve the secret
dataw = 1100 as follows:
0010
0100
1000
1100
f a′ :⊕ 1110
______
1100
(14)
5. ARCHITECTURE
5.1 Simulation and Analysis
In order to check the performance of the proposed algorithm,
we tend to use a meaningful binary logo image with 64 by
64pixels as the secret message and hide it into the carrier
image Lena with 256 by 256 pixels shown in Fig. 4(a and b).
The corresponding stego image is shown in Fig. 4(c).
The PSNR of original carrier image and stego image is 70.33.
It is indicated that the extended method has a strong
covertness. Fig. 5 illustrates the difference between the
embedding efficiency of standard matrix encoding and that of
extended algorithm with three different maximal extension
layer (respectively L = 7,15,31).
From the figure it can be seen that the embedding efficiency of
the extended algorithm is greater than that of matrix encoding
when the ratio of k to n is larger. However, when the ratio is
less than 3/7, the efficiency of extended algorithm with L = 7
starts to be lower than that of standard algorithm. When the
ratio is less than 4/15 and 5/31, the efficiency of extended
algorithm with L = 15 and 31 starts to be lower than that of
standard algorithm, respectively. Although the extended
algorithm increases the bit number of embedded secret
message, the modification for setting symbol positions affect
the embedding efficiency because these changes do not load
any secret message. Actually, the bit number of average
changes for setting m-bit symbol is m/2 bits.
For every embedding cell, there are average m/2 bits
modification used to set symbol rather than loading secret
message. With the decrease of the ratio of kto n, the length of
carrier data n becomes large which leads to the significant
reduction ofthe number of embedding cells generated by
partitioning the LSB plane of quantized DCT coefficients. In

__________________________________________________________________________________________
other words, the proportion of altered bits in carrier data is
very small. In this case, the influence of symbol modification
emerges and the cost of changing without loading effective
information cannot be neglected. Oppositely, when the ratio of
k to n is large, moresecret data can be embedded and a large
number of changes exist inside the cover data. The influence
of symbol modificationis relatively little. Although we can
improve the embedding efficiency by increasing the maximal
extension layer L, the increase of L will result in the increase
of length of symbol which has a negative impact on the
efficiency. Thus, we should findout the optimal tradeoff to
confirm the coding mode (1, n, k, L) according to the size and
type of secret message. Taking the coding mode (1,3,2, L) for
example, a new logo image illustrated in Fig. 6(a) will be
embedded into the test image Lena using the extended
scheme. It can be seen from Fig. 6(b) that the values of
embedding efficiency vary in according to the maximumof
extension layer L when n and k are specific. In this case, the
maximum of embedding efficiency is obtained when L =
15.When L is less than 15, the secret bits are not extended
adequately. Contrarily, a larger L leads to the increase of the
length of symbol which has a negative impact on the
embedding efficiency. Obviously, the coding mode (1,3,2,15)
is the best choice in this instance.
Fig -4: The performance of the extended algorithm
In matrix encoding, the length of secret message k is always
smaller than the length of carrier cell n . It means the
embedding rate is smaller than 100% for ever. However, it is
possible for the extended algorithm that the length of secret
message (k + lcrt ) is larger than the length of carrier cell
(n + m) due to the extension. Theoretically, the embedding
rate can exceed 100% and the maximum value can be close to
Rmax .
Rmax =
Ls
n+[log 2(Ls −k+1)]
,
(15)
Fig -5: The comparison of the embedding efficiency.

__________________________________________________________________________________________
Fig -6: The relationship between L and E when n and k are
specific.
Fig -7: The comparison of the embedding rate.
Where Ls denotes the length of the whole secret message. Fig.
7 shows the comparison between the embedding rate of
standard matrix encoding and that of extended algorithm with
three different maximal extension layer (respectively L =
7,15,31).
CONCLUSION
Many steganographic algorithms provide a high capability for
hidden messages however neglect the embedding efficiency.
In fact, fewer modifications might guarantee the high
covertness of the stegosystem. F5 steganography employs
matrix encoding to get higher efficiency than classic LSB
replacement or matching methods. However, with the increase
of the efficiency, the embedding rate is decreased rapidly. In
high efficiency case, a small amount of secret message can be
embedded that leads to a waste of carrier data. The extended
matrix encoding algorithm proposed in this paper uses n-layer
extension to enhance the embedding efficiency and rate
simultaneously. In the application of high ratio of kto n, the
extended algorithm can obtain a better efficiency. Moreover,
the extended algorithm makes the embedding rate exceed
100% theoretically and realizes a high capacity whilst
performing a high efficiency.
From some experimental results, e.g. in Fig. 5, it can be seen
that the embedding efficiency of improved scheme is not
always higher than F5 all the time. When the ratio of k to n
becomes small, the embedding efficiency gets decreased due
to the introduction of symbol bits. The reason of this
phenomenon is mainly that the changes for setting symbols do
not load any secret message and a random-type secret message
cannot always be extended as well (actually with the
possibility of 0.5). For the new method, this problem seems
inevitable. In spite of this, our scheme is still important and
meaningful because we also use binary images which are full
of consecutive black pixels ‘00000000. . .’ like the logos in
Figs. 4 and 6. This kind of secret message has more
opportunities to be extended with a large number of extension
layers. The experimental results demonstrate that the extended
algorithm has a high covertness, a high embedding efficiency
and a high embedding rate in the most cases. And the scheme
is effective enough to be applied to the area of covert
communication.
REFERENCES:
[01]. Simmons GJ. The prisoners’ problem and the subliminal
channel. In: Advances in cryptology: proceedings of crypto
83, New York; 1984. p. 51–67.
[02].vanSchyndel RG, Tirkel A, Osborne CF. A digital
watermark. In: Proc. of int. conf. on image processing, Austin;
November 1994. p. 86–9.
[03]. Franz E, Jerichow A, Moller S, Pfitzmann A, Stierand I.
Computer based steganography: how it works and why
therefore any restrictions oncryptography are nonsense, at
best. In: Proc. of the 1st international workshop on
information hiding, Cambridge; May 1996. p. 7–21.
[04]. Zhang Tao, Ping Xijian. A fast and effective steganalytic
technique against JSteg-like algorithms. In: Proc. 8th ACM
symp. on applied computing, Florida;March 2003. p. 307–11.
[05].Westfeld A, Pfitzmann A. Attacks on steganographic
systems. Lect Notes ComputSci 2000;1768:61–75.
[06].Provos N. Defending against statistical steganalysis. In:
Proc. of the 10th USENIX security symposium, Washington,
DC; August 2001. p. 323–35.
[07].Provos N, Honeyman P. Hide and seek: an introduction to
steganography. IEEE Secur Privacy 2003;1:32–44.
[08].Fridrich J, Goljan M, Hogea D. Attacking the OutGuess.
In: Proc. of the 3rd information hiding workshop on
multimedia and security, Juan-les-Pins; December 2002. p. 3–
6.
[09].Westfeld A. F5-a steganographic algorithm: high capacity
despite better steganalysis. Lect Notes ComputSci
2001;2137:289–302.

__________________________________________________________________________________________
[10].Sallee P. Model-based steganography. In: Proc. of the
2nd international workshop on digital watermarking, Seoul;
October 2003. p. 154–67.
[11].Sallee P. Model-based methods for steganography and
steganalysis. Int J Image Graphics 2005;5(1):167–90.
[12] Fridrich J, Goljan M, Soukal D. Perturbed quantization
steganography with wet paper codes. In: Proc. of ACM
multimedia workshop, Magdeburg; September 2004. p. 4–15.
[13] Solanki K, Sarkar A, Manjunath B. YASS: yet another
steganographic scheme that resists blind steganalysis. Lect
Notes ComputSci 2008;4567:16–31.
[14] Wang Xiangyang, Yang Yiping, Yang Hongying.
Invariant image watermarking using multi-scale Harris
detector and wavelet moments. ComputElectrEng
2010;36(1):31–44.
[15] Lu Wei, Lu Hongtao, Chung Fulai. Feature based robust
watermarking using image normalization. ComputElectrEng
2010;36(1):2–18.
[16] Lu Wei, Sun Wei, Lu Hongtao. Robust watermarking
based on DWT and nonnegative matrix factorization.
ComputElectrEng 2009;35(1):183–8.
[17] Al-Otum HA, Al-Taba’a AO. Adaptive color image
watermarking based on a modified improved pixel-wise
masking technique. ComputElectrEng 2009;35(5):673–95.
[18] Fan Li, GaoTiegang, Yang Qunting. A novel
watermarking scheme for copyright protection based on
adaptive joint image feature and visual secret sharing. Inf
Control. 2011;7(7):3679–94.
[20] Crandall R. Some notes on steganography. Posted on
steganography mailing list; 1998. http://os.inf.tu-
dresden.de/~westfeld/crandall.pdf.
[21] Wallace GK. The JPEG still picture compression
standard. IEEE Trans Consum Electron 1992;38(1):xviii–xxiv.
BIOGRAPHIES:
Borkar Bharat Sampatraoreceived his
B.E and M.E degrees in Computer
Science and Engineering from Pune
University, He is a member of ISTE,
ACM, IE. He published 3 National and 2
International papers. He presented 3
papers in national conference and 2
papers in international conference.His research interests
includeImage processing.
Pritesh Kashinath Patil his B.E. in
information Technology from North
Maharashtra University, in 2012. He is
currently doing his M.E. in Information
Technology from University of Pune.

Stegnography of high embedding efficiency by using an extended matrix encoding algorithm

Recommended

More Related Content

What's hot (17)

Viewers also liked (20)

Similar to Stegnography of high embedding efficiency by using an extended matrix encoding algorithm (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Stegnography of high embedding efficiency by using an extended matrix encoding algorithm