17-03-2014, 08:50 PM
Abstract—Inspired by works on the Markov process based
steganalysis proposed in [13], we propose a new steganalysis
technique based on the conditional probability statistics. Specifically
we focus on its performance against the F5 software. In
our experiment, we prove that the proposed technique works
as well or better than the Markov process based technique in
terms of classification accuracy on F5. Our main advantage is a
much better computational efficiency. With different number of
messages embedded, it can also be seen that the performance of
steganalyis depends on the message size embedded. This paper
includes the introduction to conditional probability features, how
the experiment works, and the discussion of the results.
I. INTRODUCTION
Steganography allows a user to hide a secret message in
such a way that an adversary will not be able to detect the
existence of the secret. Steganography can be dated back to
440 BC, where the tale of Demaratus sending a warning by
using a wax tablet and Histiaeus using a tattoo on his slave’s
shaved head were mentioned by Herodotus in The Histories
of Herodotus [11].
A steganography system can be considered defeated if an
attacker is able to prove the existence of a secret message
[9]. If a steganography system fails to disguise the embedded
information, there is no point in using it since the aim for
having a secretive communication has now been exposed.
Over the last decade a wide range of steganography techniques
have appeared in the literature. Similarly, a wide range
of steganalysis techniques were also made available, intended
to let an adversary determine whether an intercepted image
contains a secret message. In particular, a number of steganalysis
techniques based on machine learning have also emerged
[10] [13]. Such techniques tend to be blind, in the sense that
they do not assume any particular steganography algorithm
and can usually break a variety of algorithms. Other methods
that are specific to certain technique of steganography, such
as proposed in [7], are categorized as non-blind techniques.
In this paper, the focus is on the steganalysis of JPEG
steganography [2], particularly because JPEG is the most
popular image format on the Internet. Steganographic images
can be shared by employing a sending and receiving process,
or just by placing it on a web page for the receiver to browse
and extract the message without being noticed. By focusing
on steganalysis for JPEG steganography, this research should
be of benefit in the general steganalysis research domain and
also be of relevance to the real world implementation of
steganalysis, such as in digital forensic investigation.
This paper consists of four sections, and starts with a review
on estimated conditional probability features. Following that,
Section 2 presents the description of our works. Section
3 contains the findings and finally Section 4 discusses the
conclusion of the research.
steganalysis proposed in [13], we propose a new steganalysis
technique based on the conditional probability statistics. Specifically
we focus on its performance against the F5 software. In
our experiment, we prove that the proposed technique works
as well or better than the Markov process based technique in
terms of classification accuracy on F5. Our main advantage is a
much better computational efficiency. With different number of
messages embedded, it can also be seen that the performance of
steganalyis depends on the message size embedded. This paper
includes the introduction to conditional probability features, how
the experiment works, and the discussion of the results.
I. INTRODUCTION
Steganography allows a user to hide a secret message in
such a way that an adversary will not be able to detect the
existence of the secret. Steganography can be dated back to
440 BC, where the tale of Demaratus sending a warning by
using a wax tablet and Histiaeus using a tattoo on his slave’s
shaved head were mentioned by Herodotus in The Histories
of Herodotus [11].
A steganography system can be considered defeated if an
attacker is able to prove the existence of a secret message
[9]. If a steganography system fails to disguise the embedded
information, there is no point in using it since the aim for
having a secretive communication has now been exposed.
Over the last decade a wide range of steganography techniques
have appeared in the literature. Similarly, a wide range
of steganalysis techniques were also made available, intended
to let an adversary determine whether an intercepted image
contains a secret message. In particular, a number of steganalysis
techniques based on machine learning have also emerged
[10] [13]. Such techniques tend to be blind, in the sense that
they do not assume any particular steganography algorithm
and can usually break a variety of algorithms. Other methods
that are specific to certain technique of steganography, such
as proposed in [7], are categorized as non-blind techniques.
In this paper, the focus is on the steganalysis of JPEG
steganography [2], particularly because JPEG is the most
popular image format on the Internet. Steganographic images
can be shared by employing a sending and receiving process,
or just by placing it on a web page for the receiver to browse
and extract the message without being noticed. By focusing
on steganalysis for JPEG steganography, this research should
be of benefit in the general steganalysis research domain and
also be of relevance to the real world implementation of
steganalysis, such as in digital forensic investigation.
This paper consists of four sections, and starts with a review
on estimated conditional probability features. Following that,
Section 2 presents the description of our works. Section
3 contains the findings and finally Section 4 discusses the
conclusion of the research.