Steganography with Least Histogram Abnormality - Computer Network Security - page 395

Steganography with Least Histogram Abnormality

Xinpeng Zhang, Shuozhong Wang, and Kaiwen Zhang

Communication & Information Engineering, Shanghai University, Shanghai 200072, China
[e-mail],[e-mail],[e-mail]

Abstract. A novel steganographic scheme is proposed which avoids asymmetry
inherent in conventional LSB embedding techniques so that abnormality in the
image histogram is kept minimum. The proposed technique is capable of re-
sisting the χ test and RS analysis, as well as a new steganalytic method named

GPC analysis as introduced in this paper. In the described steganographic tech-
nique, a pair of mutually complementary mappings, F 1 and F − 1, is used, leading to
a balanced behavior of several statistical parameters explored by several stega-
nalytic schemes, thus improved security. Experimental results are presented to
demonstrate the effectiveness of the method.

2

1 Introduction

Digital watermarking and steganography are two major branches of information hiding
[1]. While watermarking aims to protect copyright of multimedia contents, the purpose
of steganography is to send secret messages under the cover of a carrier signal. Despite
that steganographic tools only alter the most insignificant components, they inevitably
leave detectable traces. The primary goal of attack on steganographic systems, termed
steganalysis, is to detect the presence of hidden data [2,3].
A widely used technique with low computational complexity and high insertion
capacity is LSB steganography that replaces the least significant bits of the host me-
dium with a binary sequence. Many steganalytic approaches have been developed to
attack it. The χ

analysis [4,5] detects the presence of hidden data based on the fact that
the occurrence probabilities of adjacent gray values tend to become equal after LSB
embedding. The method can also be used against other steganographic schemes such as
J-Steg in which pairs of values are swapped into each other to embed message bits. RS
steganalysis proposed by Fridrich et al. utilizes sensitive dual statistics derived from
spatial correlations [2,6]. In addition, the RQP steganalysis for color images [7] is
based on statistics of the numbers of unique colors and close-color pairs.
If the cover image was initially stored in the JPEG format, message insertion may
alter the quantization characteristics of the DCT coefficients, leaving a clear sign for
successful steganalysis [2,8]. If the DCT coefficients are modified in data embedding,
block effects [9] or histogram distortion [10] can be explored in the analysis against
such steganography techniques as used in J-Steg, OutGuess and F5 [11]. Lyu and Farid
proposed a universal higher-order statistical method capable of attacking nearly every
steganographic technique [12]. But in practice, it is difficult to perform the required
training with a large number of stego and cover images.
While the above approaches aim to detect the presence of hidden data, an active
warden attack can be performed, overwriting some insignificant contents in the cover

2