2019 IEEE 2nd International Conference on Computer and Communication Engineering Technology-CCET

Combined Algorithm of Steganography with Matrix Pattern and Pixel Value

Difference

Linna Zhou Yang Cao†

Institute of Information Technology Institute of Information Technology
University of International Relations University of International Relations
Beijing, China Beijing, China
e-mail: [email protected]

Abstract—Information hiding that different from encryption avoid the attention of third parties and being further
algorithm which needs to be encrypted and decrypted by analyzed by data mining or deep mining technology. In
complex mathematical models hides secret information in general, information hiding is such a technology that can
ordinary cover image and transmits it in open transmission achieve 5% content security like encryption algorithms, and
channel in a way that is imperceptible to human eyes. LSB
(least significant bit) algorithm is commonly used for
95% hidden security like ordinary data not being noticed.
information hiding, but it does not have good anti-detection As a sub-discipline of information security, information
ability. In this paper, a steganography combination of PVD hiding includes many aspects, steganography and digital
(pixel value difference) with MP (matrix pattern) based on the watermarking, but they are different in application. In recent
difference of pixel texture is proposed, where the former is years, steganography has been greatly developed because it
classified two portions and then the binary message is is a technology that can hide secret messages in public
embedded in the edge portion of the blue layer of the image environment and realize covert communication. Not only is
without original image in extracting, the latter algorithm it hard to be recognized by eyes, but also it can confront
generates matrix patterns representing 95 characters in the computer vision recognition. In this system, steganography
green layer of the cover image, and then revises the gray value
of the pixels in the blue layer of the cover image. The 4th- 6th
includes cover media, keys, secret information and stegano
bits of the blue layer are used for MP algorithm, and the rest media. Secret information is the content information that the
are used for PVD algorithm. The result had been proved that it sender transmits to the legitimate receiver in a secure way
has diverse embedded message forms, better detection through an unsafe transmission channel. The key is here as
resistance and high security capacity. an unnecessary part. Lastly, the stegano media is the final
form by the cover media after embedding secret information.
Keywords-data security; information hiding; steganography; That is to say, in traditional steganography system, secret
image processing; pixel value difference ; matrix pattern information is embedded into cover media by embedding
I. INTRODUCTION algorithm and then extracted by extracting algorithm on the
receiving side, among which general digital cover media
With the rapid development of computer networks, the includes graphics, images, text, audio and video, like smart
current network space environment has gradually become phones, cloud servers and video traffic are also used as
too immense and complex to solve. The network security cover media in some emerging areas. In this paper, image
issues have been attached extensive importance by industry- steganography technology is discussed.
wide and all various field people, not only professionals in In [1], an improved LSB-based algorithm is proposed,
related area with the concept of cyberspace security which uses pseudo-random sequence to modify the lsb of
presented. As we all know, the offense and defense in the the image randomly, and embeds secret information into the
network world are unequal and in the face of increasingly brightness value of the pixel according to the statistical
diversified attacks, the difficulty of defense side is harder a characteristics of the pixels of the cover image [2]. Other
lot. The network security management that strictly guard spatial algorithms take into account the characteristics of the
against deadlocks and physical isolation is widely accepted pixels of the cover image, such as the general matrix pattern
all the time, however the current network environment is algorithm [3] and the improved matrix pattern [4]. The PVD
transforming into another new normal state, which can be algorithm uses the difference of gray value between two
described as you have me and I have you. adjacent pixels in cover image as a feature with the different
It obviously be stretched to the limit if we still use such amount of secret information embedded into diverse cover
measure to solve the issues. So some experts propose image categories [5], [6]. The combined MSB and LSB
another management that focusing on evacuation, algorithm is presented with setting variable ranges
synchronizing sorting and governance. In fact, network according to the characteristics and color component values
security ultimately protects data, whether it can focus on the of each bit plane and the range is changed in the binary
data and specifically data behavior itself, which is able to secret information, in addition, the hidden information is

978-1-7281-2871-9/19/$31.00 ©2019 IEEE 6

Authorized licensed use limited to: University College London. Downloaded on July 04,2020 at 05:05:29 UTC from IEEE Xplore. Restrictions apply.
scrambled and randomly embedded into the cover [7]. Some the corresponding pixel values are qi and qi+1 (i is the ith
image processing and steganalysis techniques using block traversed). The difference is calculated as
machine learning and deep learning have gradually emerged. d =| qi − qi +1 | , d  (0,255). Over the threshold determined
In [8], a steganalysis algorithm against spatial are selected for embedding. Then put d into a continuous
steganography without manual intervention based on deep range [0,255] and marked as Rk,, among which k denotes the
learning is proposed with the model framework consists of a d is the k-th part of R. u k and l k are the upper and lower
high-pass filter, two convolution layers and two full-
connection layers. Some scholars have also put forward a bounds of R k , so d = u k − l k + 1 ,d is chosen as the integer
general steganalysis algorithm based on supervised learning power of 2 as far as possible so that it can be extracted
and unsupervised learning [9]. Steganalysis and better. Generally, a equation is reached as
steganography are antagonistic but their development d = uk − lk + 1 = 2n , n is the integer. Because this algorithm
promotes to each other. Some state of the art steganography
hides information by transforming secret information into
algorithms based on machine learning or deep learning are
binary information (such as 0 and 1 bit sequence), secret
growing increasingly. In [10], a steganography without
information requires to be embedded bit by bit. The number
embedding (SWE) is proposed, focusing on the relation
between secret information and covers to prevent of bits denoted as n and n = log (2uk −lk +1) .Then the n bits
steganalysis detection based on machine learning [11]. In secret stream value bk need to be calculated and the new
SWE, cover image itself represents secret information, difference value d ' = bk + lk replaces the original d [5], and
because of abundant information, such as gray value, color,
texture, edge, contour and advanced semantics in digital the new gray value f((qi ,qi + 1 ),m) = (qi' , qi' +1 ) of stegano
image, it is very important to process and use these image is obtained by (1):
information reasonably and establish mapping between
them and secret possibly, and a lot of research has been  qi − m +1,qi+1 + m −1, where d is odd, m is odd
done [12-16]. In this paper, we combine MP and PVD in
' '
 m2−1 2
m +1
spatial domain to form a new steganography algorithm. The (qi ,qi +1 ) = qi − , qi +1 + , where d is even,m is odd ()
next part of the paper is divided into the follows: section II 2 2
 m m
is describes PVD algorithm work, the MP algorithm is  qi − , qi +1 + , where m is even
2 2
presented in section III, In section IV, embedding and
extracting process of combination algorithm are shown, where m = d ' − d .However, the new gray value (qi' , qi' +1 )
Section V and section VI illustrate simulation results, may exceed the range of [0,255], which will easily
conclusion and future work, respectively. detectable. Therefore, before embedding, it is necessary to
II. PVD ALGORITHM select a suitable block by some ways. We can calculate the
f (( qi , qi +1 ), m) , if the result is within the range of [0, 255],
PVD, known as pixel value difference, is an efficient
steganography based on the imperceptible capacity of the it indicates that it can be used as a cover block to be
human eye to hide data with using the difference of gray embedded, otherwise it will be abandoned. Similarly, this
value between two adjacent pixel blocks in the cover image test method can also be applied in extracting.
that divided into smooth and rough by the selected threshold When extracting secret information, the receiver first
value. Then, different amounts of data are embedded in traverses the stegano image with the same order as
different categories in the light of different smoothness. This embedding stage before, assuming that the pixel values of
method provides simpler and more effective imperceptible the first traversed block are q1* and q2* , the difference of
way than LSB method. In addition, the advantages of this pixel values is d * =| q1* − q2* | and calculating
method that the original image is not needed when secret
message is recovered and achieves confidentiality that f((q1* , q2* ), uk *
− d ) with (1). If the value exceeds the range
preventing the access and tampering from illegal users of [0, 255], there is no secret information in the block,
according to a traversing cover image order provided by the otherwise the binary sequence value of secret information b k
pseudo-random generator is also outstanding. is calculated by equation (2):
In the embedding stage, the adjacent two pixels of cover
image is first divided into edge regions and smooth regions bk = d * − lk ()
according to the difference. The larger the difference, the
rougher the region is, otherwise it is smooth region. III. MP ALGORITHM
Experiments have proved that information hiding in the Before this algorithm, there is a static matrix pattern
edge region has better privacy and is not easy to detect. The algorithm that needs artificially select the matrix pattern to
cover image is traversed according to the way determined represent the corresponding characters, which is similar to
by the pseudo-random generator, and two non-overlapping common code book between the sender and the receiver in
blocks are constructed, which called pi and pi+1 respectively, encryption algorithm. But this work is time-consuming and

Authorized licensed use limited to: University College London. Downloaded on July 04,2020 at 05:05:29 UTC from IEEE Xplore. Restrictions apply.
does not take into account the difference of the pixel blocks exceeds 120, or the same as the matrix before, the pointers
in the cover image. In [3], a more flexible and anti- all will move a column to the right. When lines run out, the
detectable matrix pattern algorithm is proposed, which uses pointer moves down three rows and continues from the
the difference of pixel in a small range to construct dynamic leftmost until 95 different matrix patterns generate, which
matrix patterns to represent different characters. The can be denoted as A'{a1' , a'2 , a3' , a'4 ,..., ai' } ,i=1,2,3,... 95, A ' is
following contents further elaborate the process of matrix
a matrix composed of 95 matrices.
pattern algorithm.
Firstly, a BxB Block suitable for generating matrix TABLE II. THE FIRST ‘STEGANO-MP’
pattern and embedding secret information is selected in the
cover image, then its size, location and the size of the matrix
pattern bxb are determined, among which the size of BxB
and bxb have been decided by the results that produce the
best experimental effect, but the location of Block is
obtained by the pseudo-random generator. By the way, the
seed of pseudo-random and the sizes of matrix pattern and
Block will be embedded by this algorithm. Then, a matrix
pattern representing 95 characters including 26 English
letters(big and small), 10 digits, 32 keyboard non-
alphanumeric symbols and one message ending character
can be generated in BxB Block. Assuming that the size of In the stage of embedding secret information, the pixel
the Block is 60x60 and matrix pattern is 3x3, respectively. values are converted to RGB mode, and the secret
The RGB image is switched to brightness mode through the information is embedded into the unused positions which
following equation: segmented as B{b1 , b 2 , b 3 , b 4 ,..., b i } ,i = 1, 2, 3,...,k.
Intensity ( p( R, G, B)) = ( R + G + B ) / 3 () Starting from the leftmost side of B, the first mp
a i' corresponding to the first character of secret information
In (3), R, G and B represent the gray values in the three
layers of red, green and blue of the selected Block. plus b 1 get b1' , specifically, the first row of the stegano
Taking the generation of matrix pattern of character ‘A’ matrix pattern equals the first row of b 1 , the second row of
as an example: from the selected Block top left select a 3x3
block and ignore the first three bits of each pixel, because b1' equals the first row of b1' plus the second line of a i' , and
they are often detected by the LSB-steganalysis method.. the third row of b1' equals the second row of b1' plus the
Then set the value of first row in 3x3 mp to zero, and the
third row of a i' . According to the same process, the stegano
second row value of the mp equal the second row value
minus the first row value ,the third row value of the mp matrix pattern have completed, denoting as
equal the third row value minus the second row value. The B'{b1' , b '2 , b3' , b '4 ,..., bi' } , i = 1, 2, 3,..., k. The pseudocode is
pseudocode generating mp representing ‘A’ is shown in shown in Table 2.
Table 1: For extracting secret information, according to what has
TABLE I. THE MP OF ‘A’
been said before, the seed of random sequence, the size of
Block and matrix pattern, the location of embedded secret
information are extracted first. Then converting to the
brightness mode according to (3) and the first three bits of
each pixel brightness value are discarded. The second row
of stegano mp is subtracted from the first row as the secret
character’s second row, the third row stegano mp is
subtracted from the second row as the secret character’s
third row, and its first row is set to zero. In this way, all mps
representing secret information are completely extracted
until representing message ending character appears.

Next, the pointer of the first pattern moves a column to IV. PROPOSED METHOD
the right, and a matrix pattern representing the next MP algorithm only embed text message into cover
character is generated by the same process. The pointer image, while PVD algorithm can convert any message such
moves a column until it complete the assignment of the last as signal, audio, video, numerical value, text into binary
character in this line. Moreover, if the values in the matrix form and these can be embedded in cover image. Obviously,
pattern generated in this process are all zero, one of values MP algorithm can express any letter (case-sensitive),

Authorized licensed use limited to: University College London. Downloaded on July 04,2020 at 05:05:29 UTC from IEEE Xplore. Restrictions apply.
number, and other symbols, which make it possible to hide B. Extracting Secret
computer codes. In this algorithm, we use the green layer of 1) The size of the Block and the mp, the seeds are
cover image after ignoring the first three bits to generate extracted by the MP extraction algorithm from the Block.
matrix patterns representing 95 characters. The fourth to 2) After determining the number of seeds, the pseudo-
sixth bits of the blue layer are embedded secret with MP random generator is used for generating the same
algorithm and the blue layer of the edge is embedded with embedding order as the sender.
PVD algorithm. The combination of the two algorithms has 3) When the first Block is selected, the brightness value
good anti-detectability and high steganography capacity. is calculated by equation (3), the text information is
extracted by MP extraction algorithm until the the mp of
text information terminator appears or the Blocks are used
up.
4) After extracting the text information, binary
information is extracted by PVD extraction algorithm.
V. SIMULATION RESULT
In this section, the simulation with 13 pictures using the
proposed algorithm proves that we get improved embedding
capacity and more valid anti-detection ability through some
performance indexes and result clearly shown in Table III .
TABLE III. SIMULATION RESULT OF 13 PICTURE

Figure 1. A flow chart of combination method.

A. Embedding Secret
The two algorithm implementations in detail have given
in section II and section III, respectively. The process of
combination of the two algorithms is as follows. The flow
chart is given in Figure 1.
1) Determining the size of BxB Block, b1xb2 matrix
pattern, and the seed of pseudo-random generator.
Embedding these information in the 3x2 mp and 64x64
Block with MP algorithm. By the way, experiments have
shown these sizes can achieve the best security capacity,
anti-attack ability, and it is sufficiently large to hide secret
information [3]. Figure 2. Histogram of gray level for Lena image.
2) The selected cover image is divided into several non-
overlapping Blocks according to 1). Finding the location of Our algorithm does imperceptible changes in the
the Blocks according to the generated numbers by the histogram of gray level for stegano image, as shown in
random generator and ready to be embedded. Figure 2, we compare source cover image with embedding
3) Converting the value of the image Block into secret stegano image. As we can clearly see, the two pictures
brightness mode, and generating "matrix pattern" in Figure 2 are basically the same, thus our algorithm can
representing 95 characters after ignoring the first three bits successfully resist this kind of attack.
in the green layer. X 2 analysis can determine whether there is secret
4) Embedding text message into the blue layer of cover information in image, but it requires that the secret
image with MP method until the text embedded completely, information be embedded 100% in all the pixels of the cover
otherwise go to step 5). image. In this research, except for embedding the massage in
5) Converting text message into binary message. the edge part of the cover image with PVD, the rest are all
embedded in the whole image with MP algorithm, and the
6) The binary message is embedded in the blue layer of
the edge part of the cover image with PVD algorithm and resistance to X 2 attack is enhanced, as shown in the Figure 3.
obtain a stegano image.

Authorized licensed use limited to: University College London. Downloaded on July 04,2020 at 05:05:29 UTC from IEEE Xplore. Restrictions apply.
 [2] Thangadurai K，Sudha Devin G. An analysis of lsb based image
steganography techniques[C]//Int Conf on Computer Communication
and Informatics. Piscataw，NJ：IEEE，2014
[3] Amir Farhad Nilizadeh，Ahmad Reza Naghsh Nilchi. Steganography
on RGB Images based on a”matrix pattern” using random
blocks[J/OL]. Modern Education and Computer Science Press ，
2013，4(2)：8-18. http://www.mecs-press.org/
[4] Amirfarhad Nilizadeh ，Wojciech Mazurczyk ，Cliff Zou， et al.
Information hiding in RGB images using an improved matrix pattern
approach[C]// 2017 IEEE Conf on Computer Vision and Pattern
Recognition Workshops. Piscataway ， NJ ： IEEE ， 2017 ： 1407-
1415
2 [5] Wu Dachun, Tsai Wenhsiang. A steganographic method for images
Figure 3. Simulations results from X attack on Lena. by pixel-value differencing[J]. Pattern Recognition Letters，2002，
24 (9)：1613–1626
VI. CONCLUSION AND FUTURE WORK [6] Zaker Nazanin，Hamzeh Ali， Seraj Dean Katebi，et al. Improving
In this paper, a combined algorithm of steganography security of pixel value differencing steganographic method[C]// Int
Conf on New Technology. Piscataway，NJ：IEEE，2009
with matrix pattern and pixel value difference is proposed.
[7] Yuan Ziqian ， Guan Zijie ， Feng Hao. An improved information
We elaborate on the embedding and extraction process of hiding algorithm based on image[C]// IEEE Int Conf on Software.
PVD algorithm, MP algorithm, and the combination Piscataw，NJ：IEEE，2017：169-172
algorithm. We apply this method to 13 pictures with [8] Kim Donghyun, Lee Haeyeoun.Deep learning-based steganalysis
simulation results shown that it has better anti-detection against spatial domain steganography. 2017 European Conf on
Electrical Engineering and Computer Sci,2017:1-3
ability and high security steganography capacity.
[9] F. K GOMIS,M.S Camara,Idy Diop,S.M Farssi,et al.Multiple linear
Furthermore, as machine learning and deep learning regression for universal steganalysis of images .Centre d’Excellence
technologies become more and more mature in the field of Africain en Mathématiques,Informatique et TIC.IEEE,2018.
image processing, some steganalysis achievement based on [10] Barni M. Steganography in digital media: principles, algorithms, and
these technologies have developed rapidly, but few research applications (Fridrich, J. 2010) book reviews[J]. IEEE Signal Proc
results in steganography. In the next work, we will try add Magazine，2011，28(05)：142–144
some mature deep learning frameworks, such as DBSCAN to [11] Fridrich J. Steganography in digital media: principles, algorithms, and
applications[M]. Cambridge：Cambridge University Press，2009：
make efforts for image steganography. 1–20
[12] Zhou Zhili，Cao Yi，Sun Xingming. Coverless information hiding
ACKNOWLEDGMENT based on bag-of-words model of image[J]. Journal of Applied
The work is supported by the National Natural Science Science- Electronics and Information Engineering，2016， 34(05)：
527–536
Foundation of China (NSFC) under Grant no.U1536207, the
[13] Zhou Zhili ， Sun Huiyu, Harit Rohan ， et al. Coverless image
National Key Research and Development Program of China steganography without embedding[C/OL]// Int Conf on Cloud
under no.2016QY08D1600 and the National Key Research Computing and Security. Berlin ： Springer ， 2015 ： 123–132.
and Development Project of China under https://link.springer.com/chapter/10.1007%2F978-3-319-27051-7_11
no.2016YFB0801405.We would like to thank Dr. Xin [14] Zheng Shuili ， Wang Liang ， Ling Baohong ， et al. Coverless
Tang’s useful suggestions and comments. information hiding based on robust image hashing[C]// Int Conf on
Intelligent Computing Methodologies. Switzerland ： Springer ，
2017：536–547
REFERENCES
[15] Xu Jiayi，Mao Xiaoyang，Jin Xiaogang，et al. Hidden message in a
[1] Luo Weiqi，Huang Fangjun，Huang Jinwu. Edge adaptive image deformation-based texture[J]. The Visual Computer ， 2015 ，
steganography based on LSB matching revisited[J]. IEEE Trans on 31(12)：1653–1669.
Information Forensics and Security，2010，5(2)：201-214
[16] Wu Kuochen ，Wang Chungming. Steganography using reversible
texture synthesis[J]. IEEE Trans Image Processing，2015，24(01)：
130–139.

10

Authorized licensed use limited to: University College London. Downloaded on July 04,2020 at 05:05:29 UTC from IEEE Xplore. Restrictions apply.