Analysis of image steganalysis techniques to defend against statistical attacks – a survey

IJRET: International Journal of Research in Engineering and Technology ISSN: 2319-1163
__________________________________________________________________________________________
Volume: 01 Issue: 02 | Oct-2012, Available @ http://www.ijret.org 148
ANALYSIS OF IMAGE STEGANALYSIS TECHNIQUES TO DEFEND
AGAINST STATISTICAL ATTACKS – A SURVEY
Usha B.A1
, N K Srinath2
, N K Cauvery3
1, 2, 3
Asst Professor, Prof and HOD, Professor, Dept of CSE, R V College of Engineering, Bangalore, India
ushaba@rvce.edu.in, srinathnk@rvce.edu.in, cauverynk@rvce.edu.in
Abstract
Steganography is the art concealing information to transmit it in such a way that nobody but the intended receiver knows the existence
of the message. Steganalysis techniques work on eliminating suspicion about the existence of a message. If suspicion is raised, then
the message cannot be passed covertly. One of the ways to detect the hidden message is to view the statistical properties of the image
or medium in which the message is hidden. This is called a statistical attack. In this paper, we explain the nature of such attacks and
present our conclusions based on reviews of existing methods of defense against statistical attacks.
Keywords– Steganalysis, Steganography, statistical attacks, JPEG images, OutGuess
-----------------------------------------------------------------------***-----------------------------------------------------------------------
1. INTRODUCTION
Steganography comes from Greek words meaning "covered
writing". Over the ages, it has developed into an art of hiding
the very existence of a message when it is being relayed. Any
code that can be represented digitally can act as a carrier for a
message. This could be text, an image, a video or any other
such “innocent looking” media. A message is the information
hidden in anything that can be embedded into a bit stream like
plain text, cipher text or even another image. A stego-carrier is
the name given to the cover carrier and the embedded message
together. A stegokey is any additional hidden information, such
as a password, required to be able to send a message. A
possible formula of the process may be represented as: [1]
Cover medium + embedded message + stegokey = stego-
medium
1.1 Methods Used
Due to their omnipresence in the World Wide Web, images
provide excellent carriers for hidden information. Many
different techniques have been introduced till date [2]. Each of
these can be used to test detection properties and robustness in
the effort to destroy or disable the embedded message. These
techniques can be broadly categorized into two groups: those in
the Image Domain and those in the Transform Domain.
i) Image Domain tools make use of bit-wise methods that apply
least significant bit (LSB) insertion and noise manipulation.
These “simple system” approaches in steganography [3] set the
least significant bits of image pixels equal to those of the
message text.
The tools used in this group include StegoDos, S-Tools, Hide
and Seek, Hide4PGP and Steganos among others. There is no
loss in the image formats used in such steganography and it
allows data to be directly manipulated and recovered. This
approach also comprises including additional components such
as masks or image objects to watermark an image.
ii) The transform domain tools involve manipulation of
algorithms and image transforms such as discrete cosine
transformation (DCT) and wavelet transformation. These
methods hide messages in more significant areas of the
covering image medium and hence may end up manipulating
image properties such as brightness. Watermarking tools
belong to this category. Software that implements this approach
includes PictureMarc, JK-PGS, SysCop, and SureSign. This
approach is more robust than bit-wise techniques but there
exists a tradeoff between this robustness and the amount of
information that can feasibly be added to the image. [1]
JPEG images use the Discrete Cosine Transform (DCT) to
achieve image compression. The compressed data is stored as
integers; however, if we need to quantize the data to encode a
message, all the calculations required involve floating point
data. Information is hidden in the JPEG image by modulating
the rounding choices either up or down in the DCT coefficients.
Detection of such an embedded message would seem to be
quite difficult. In this rounding off, errors may occur and this
leads to the losses in this method. The tool Jpeg-Jsteg is a
steganography tool that makes use of this property. In
steganography when we divide the image into 8x8 sub-images,
the boundaries of these sub-images may become visible,
leading to a disjoint image. This is called blocking artifact and
DCT is used to minimize it. [4]

__________________________________________________________________________________________
We can also use tools whose approach combines image and
transform domain tools. These may include approaches like
patchwork and masking which make the hidden information
seem redundant. [3] Hence the steganography is successful and
independent of effects like cropping and rotation. Patchwork
uses a pseudo-random technique to select multiple patches of
an image for marking with a watermark so that even if one is
lost, the entire message is not lost. Masks use image domain
tactics since they are just an extra component or image object.
They also use transform domain tools for adjusting image
properties or transforming them.
1.2 Steganalytic Detection
For a person applying Steganalysis to determine the presence of
a message, the method he will use is called a steganalytic
attack. There are five main types of attacks possible, namely
stego-only, known cover, known message, chosen stego, and
chosen message. A stego-only attack is one where only the
stego-medium is available for analysis. If the "original" cover-
media and stego-media are both available, then a known cover
attack is available. A known message attack is used when the
hidden message is revealed at some later date, and an attacker
may attempt to analyze the stego-media for future attacks. The
chosen stego attack is one where the steganography algorithm
and stego-media are known. A chosen message attack is one
where the steganalyst generates stego-media from some
steganography tool or algorithm and a known message so that
he can find out which tools and algorithms are used based on
the patterns he can detect.
2. STATISTICAL ATTACKS
2.1. Characteristics Of A Statistical Attack
While embedding the message in a medium, there are certain
statistical properties of the medium that need to be maintained.
If not maintained, they can reveal the existence of a hidden
message. For example if the medium is an image, examples of
such properties would include PoVs. A PoV or pair of values is
a tuple (2i, 2i +1) that indicates that the bit denoted by 2i is
transformed onto the bit 2i+1 after encoding the message within
the image. These PoVs are automatically generated when the
image is encoded and can be stored. Then, for the kth pixel, the
frequency histogram of the image, denoted by Yk will change
and then the presence of the message will become apparent.
Hence we need to maintain that Y2i+Y2i+1 remains a constant.
If not, this type of attack is called a statistical attack.
The main types of statistical analysis include histographic, chi-
square, generalized chi-square and pairs analysis.
Histographic analysis on JPEG sequential and pseudo-random
embedding type stegosystems, such as JSteg and Outguess 0.1
can effectively estimate the length of the message embedded
and it is based on the loss of histogram symmetry after
embedding. χ2 makes use of the generalized Euler Gamma
function and has a rigorous mathematical proof.
The generalised χ2 attack does not calculate an estimation of
the message length and can be sometimes wrong if the message
has a significant difference in the number of zeros compared to
ones.
Pairs analysis was designed specifically for GIF images but
works well for greyscale BMP images as well and can estimate
the length of the message embedded. [5]
3. CURRENT APPROACHES IN STEGANALYSIS
At present, there are two approaches to the problem of
steganalysis, one where there exists a specific steganographic
technique for a specific steganographic algorithm or one where
the technique is independent of it. The success of these
algorithms depends on its capacity to detect the presence of an
image. Decoding of a message can be cumbersome if strong
cryptographic tools have been used. If the individual is able to
detect the presence of a message, then the steganographic
system is said to be insecure [7] and we need to find ways to
ensure secure transmission of secret messages.
Here, the authors discuss the possibilities of deterministic and
non-deterministic approaches in steganography. The
deterministic method proves to be ineffective if both the stego
and the cover are known to a third person that wants to break
into the system. In [7], we see that it is impossible to have
secure data in such cases. Thus, indeterminism is introduced in
the embedding operation. This results in randomness each time
the process is computed thus introducing uncertainty. Thus, if
the attacker is aware of the stego and the cover (S and C
respectively), then when S is known, C is uncertain, such that
H(C/S) > 0 where H denotes the entropy. To ensure this, the
authors in [7] introduce another variable CS from which the
actual cover is selected. They assume that the steganographic
function, CS and the stego are publicly known whereas the key
and cover are unknown to the attackers. They prove that the
cover must contain an uncertainty for the attacker to allow
secure steganography between sender and recipient. Thus, to
ensure security, the embedding process must be split into two:
non - deterministic and deterministic parts, which are
indistinguishable to the attacker.
For instance, an image taken using a camera can easily portray
the area where the image was captured and the features of the
camera. But, the position of the camera or the direction of the
camera while capturing the image cannot be predicted
accurately. Thus, the attacker is unaware of the image details
and cannot distinguish between an original image and a
steganographic one. Thus, we can say that preprocessing i.e.
positioning of the camera initiates randomness into the cover
data.
Another method to ensure secure data sending illustrates the
idea of a trusted domain. In [7], we see that ISDN allows secure

__________________________________________________________________________________________
communication with steganography with the aid of bit
transparent transportation of digital data. Here, the attacker fails
to encode the secret message without the key even if he knows
the source and the embedding function. The characteristic of
the output (stego or cover) cannot be determined from outside
in this method.
Information hiding begins by modifying the redundant bits,
which can change the statistical properties of the cover. For
instance, equal likelihood of 1’s and 0’s occurs. But, redundant
data tends to converge towards one of them. Embedding data
thus weakens the correlation. [9] Thus, for secure
steganography, the key and the actual cover must remain
unknown to the attacker. We also need to apply additional
transforms to the redundant data, which correct measurable
deviations in the embedding process. This is possible by
concealing the key with a symmetric cryptosystem. The
embedding of the message in the image must also allow for
randomness to enhance security. One must also be aware of the
preprocessing techniques for effective steganographic
implementation of messages.
4. COMMONLY USED STEGANOGRAPHY
ALGORITHMS
Based on the images they operate on, the algorithms can be
classified into three types – raw images, palette based and
JPEG images.
4.1 Raw Images
We understand that the simple LSB (Least Significant Bit)
embedding method operates on raw images, where the message
is embedded in a subset of the LSB plane of the image usually
after encryption [6]. As discussed earlier, this introduces
partitioning of the image into PoVs. Since the values are
mapped, it provides statistical flattening of the distribution of
values.
However, the major drawback of this technique is the length
constraint as it embedding can be detected only when the length
is comparable to the number of pixels in the image. Analysis
from [6] shows that only with prior knowledge of where the
message is placed, messages of shorter length can be detected
which is too strong an assumption to make.
4.2 Palette Based Images
In case of palette based images [6], we see that by observing
the palette tables in GIF images and the anomalies caused by
stego tools that perform LSB embedding in GIF images, we can
distinguish between stego and cover images.
The author emphasizes that in order to reduce the distortion
caused by embedding as proposed by EzStego in [10], the
colour pallet is sorted such that the colour differences between
consecutive colors is minimized. The message bits are then
embedded in the LSB of the indices. Since similar pixels,
which can modify due to the embedding process, get mapped to
the neighboring colors in the palette, visual artifacts are
minimal and hard to observe.
4.3 JPEG Images
JPEG format is one of the most widely used formats these days
and is subjected to constant research in terms of steganalysis
and steganography. One of the most frequently used algorithms
is the OutGuess Embedding Program proposed by Fridrich.
It embeds the data in LSB of the DCT coefficients randomly,
leaving some coefficients unchanged. In order to preserve the
original histogram of DCT coefficients, the remaining
coefficients are adjusted. Thus, the method that involves
estimation of the original histogram proves to be ineffective.
We discuss the OutGuess program again briefly. [6]
5. STATISTICAL EMBEDDING
In [9], the author discusses embedding techniques that
distribute the hidden message uniformly across the image.
Provos [9] explains that this process can be split into two stages
– Identification of redundant bits and selection of bits where the
data is to be placed.
Provos [9] emphasizes that the image format plays an important
role in identifying the redundant bits. He also discusses the
importance of pseudo random number generator (PRNG) that
introduces randomness. For bit selection, a cipher is used. We
understand this process from [9]. We realize that since the
PRNG has a secret key, it is impossible to detect the message
without the key.
6. STATISTICAL DETECTION
By determining that an image’s statistical properties that
deviate from a norm, we can figure out whether an image has
been modified steganographically. Certain tests do this by
measuring the entropy of the redundant data, which is higher if
the image has a hidden message. [8]
We realize that DCT coefficients play a significant role in the
embedding process as observed by the authors in [8]. The
different steganographic systems in use include -JSteg, JSteg-
Shell and JPHide and OutGuess. These are popular systems for
JPEG images. These systems utilize a form of least- significant
bit embedding and are detectable by statistical analysis except
for the OutGuess. The authors have discussed various aspects
on these systems in [8].
[8] In the JSteg system, the DCT coefficients are modified
continuously from the beginning of the image. It does not
support encryption and has no random bit selection. Here, the
first five bits of the header contain the size of the length field in
bits and the remaining ones express the size of the embedded

__________________________________________________________________________________________
data. [8] In case of JPHide, detecting content is more difficult a
task as the DCT coefficients isn’t selected continuously from
the beginning unlike the first system JSteg. It uses a fixed table
that defines classes of DCT coefficients to determine the order
in which the coefficients are to be modified. In this technique,
the next class is chosen only after the coefficients in the current
class are used. This ensures information hiding in the class even
after the entire message has been embedded. This ensures
enhanced security as the all the DCT coefficients involved are
modified even if the message is only 8 bits long (approx. 5000
co-eff.). Using this technique, we can modify the second least
significant bits of the DCT coefficients apart from the usual
least significant bits.
On the other hand, we find the OutGuess technique to be more
efficient than the previous two discussed in [8]. This chooses
the DCT coefficients with a pseudo-random number generator.
A user-supplied pass phrase initializes a stream cipher (which
encrypts the content) and a pseudo-random number generator,
both based on RC4. The authors observe that the newer version
OutGuess 0.2 preserves statistical properties unlike the version
0.13b which is vulnerable. When the OutGuess 0.2 technique is
used, the χ2-test discussed earlier fails. In [9], Provos
demonstrates how the OutGuess program preserves the
statistical properties of the image. Thus, we can preserve data
in the image and ensure effective communication without the
attacker knowing the content.
7. NEED FOR EFFECTIVE DETECTION AND ITS
APPLICATIONS
Steganography is a growing field with serious applications in
the world today. There is no way to control the huge number or
type of images being sent over the Internet and hence we have
no way to individually inspect each of them to ascertain the
presence of a hidden message. This could lead to dire
consequences like terrorist activities, financial scams or even
large-scale brainwashing. For example, in 2001, US officials
stated that they have suspicions that terrorists communicate
using steganography via the Internet.
“Hidden in the X-rated pictures on several pornographic Web
sites and the posted comments on sports chat rooms may lie the
encrypted blueprints of the next terrorist attack against the
United States or its allies. It sounds far fetched, but U.S.
officials and experts say it's the latest method of
communication being used by Osama bin Laden and his
associates to outfox law enforcement. Bin Laden, indicted in
the bombing in 1998 of two U.S. embassies in East Africa, and
others are hiding maps and photographs of terrorist targets and
posting instructions for terrorist activities on sports chat rooms,
pornographic bulletin boards and other Web sites, U.S. and
foreign officials say.” Such situations require identical counter
attacks and hence the governments of various countries, as well
as anti terrorism organizations, intelligence and espionage
bureaus need to be able to pass steganographically-encoded
information without its being threatened by statistical attacks.
Even to send nuclear, defence weapons, military tactics,
espionage and foreign strategy related messages; defence
against statistical attack is definitely the need of the hour.
CONCLUSIONS
In this paper we have recorded and summarized the current
scenario in the highly stimulating field of Steganalysis. Defense
against statistical attacks in Steganalysis holds great potential
for future research considering its vast scope and extremely
important applications. The methods used in status quo are
sufficiently advanced and can provide suitable defence against
current attacks. However, with growing awareness about
steganography and the abundance of transmission media for
images and hidden information, it can only be predicted that
methods of attack will only become more sophisticated. In such
a scenario we need to constantly observe new forms of attack
and keep coming up with new techniques of defense against
them.
REFERENCES
[1] Johnson, N.F., Jajodia, S., Steganalysis of Images Created
using Current Steganography Software, Workshop on
Information Hiding Proceedings, Portland, Oregon, USA, 15 -
17 April 1998.
[2] Johnson, N.F., Jajodia, S: Exploring Steganography: Seeing
the Unseen, IEEE Computer, February (1998) 26-34
[3] Anderson, R., Petitcolas, F.: On the Limits of
Steganography, IEEE Journal on Selected Areas in
Communications, Vol. 16, No. 4, May
[4] Gonzalez, R.C., Woods, R.E.: Digital Image Processing.
Addison-Wesley. Reading, MA, (1992)
[5]http://www.computing.surrey.ac.uk/personal/st/H.Schaathun
/projects/Past/leivaditis.pdf
[6] Mehdi Kharrazi, Husrev T. Sencar, and Nasir Memon,
Image Steganography: Concepts and Practice, WSPC Lecture
Notes, April 2004
[7] J.Zöllner, H.Federrath, H.Klimant, A.Pfitzmann,
R.Piotraschke, A.Westfeld, G.Wicke, G.Wolf, Modeling the
security of steganographic systems, Proceedings, 2nd workshop
on Information Hiding, April 1998, Portland
[8] Niels Provos, Peter Honeyman, Detecting Steganographic
Content on the Internet, Center for Information Technology
Integration, University of Michigan
[9] Niels Provos, Defending Against Statistical Steganalysis,
CITI Technical Report 01-4, Center for Information
Technology Integration, University of Michigan
[10]R. Machado, “Ezstego,” http://www.stego.com, 2001.

Analysis of image steganalysis techniques to defend against statistical attacks – a survey

More Related Content

What's hot (16)

Viewers also liked (20)

Similar to Analysis of image steganalysis techniques to defend against statistical attacks – a survey (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Analysis of image steganalysis techniques to defend against statistical attacks – a survey