06-10-2012, 12:04 PM
VQ Applications in Steganographic Data Hiding Upon Multimedia Images
VQ Applications.pdf (Size: 686.55 KB / Downloads: 28)
Abstract
Data hiding is one of the most important techniques
to achieve better data and communication protection by hiding
information into a media carrier. It provides a secure method
to distribute data through a public and open channel. Data
hiding for vector quantization (VQ)-based images focuses on
the problem of embedding secret data into a cover VQ-based
image to achieve secret communication and data protection.
This paper provides a state-of-the-art review and comparison
of the different existing data-hiding methods for VQ-based
images. In this paper, we classify VQ-based data-hiding methods
into four nonoverlapping groups according to their reversibility
and output formats, introduce the details of the representative
methods, summarize the features of the representative methods,
and compare the performance of the representative methods
using peak signal-to-noise ratio, capacity of secret data, and bit
rate. Our paper shows that an irreversible method is very likely
a VQ-based data-hiding method that produces a stego-image as
its output, and it can embed more secret data than a reversible
method. Nonstandard encoding methods (e.g., joint neighboring
coding) are becoming popular in reversible data hiding since
they can increase the capacity for embedding the secret data.
Some methods with high compression rate, such as the searchorder
coding-based methods, may reduce the compression rate
in return for the capacity for the secret data.
Introduction
THE RAPID growth of the Internet facilitates data sharing
among people. The Internet provides abundance of data
that is necessary to modern life, and accessing the Internet has
become part of many people’s daily life. However, the Internet
is public, open, and important to many people. Many security
problems may occur at any time, and they are becoming more
critical than ever. The major problem of publicity and openness
is that malicious people can attack or steal personal data easily,
even though the victims may put many efforts on privacy protection.
It has been reported that hackers’ attacks are becoming
more common and fiercer [1], [2]. This phenomenon drives
the demands for better data and communication protection
mechanisms.
VQ-Based Image Coding System
Data hiding for digital image is becoming prevalent because
digital image is the most popular multimedia type in the
Internet. Many data-hiding methods have been proposed for
different image formats in the literature [11]. VQ [23], an image
format that provides high compression rate and high image
quality, is practical for sharing and distributing image over
the Internet. Therefore, a VQ image is an appropriate media
carrier to embed secret data. VQ is a lossy data compression
method based on the principle of block coding. A VQ image is
partitioned into a series of nonoverlapping rectangular blocks,
each of which has size of k. Then, each block maps to a finite
subset of the VQ blocks Y = {yi|i = 1, 2, . . . , N}, where yi
is called a codeword, set Y is the codebook generated by the
Linde–Buzo–Gray clustering algorithm, and N is the size of
set Y.
Reversible Data Hiding Using VQ-Based
Codes as Outputs
In 2006, Chang and Lin [26] proposed a reversible embedding
method, which uses one codeword to represent an
image block. The method has good compression rate and yet
produces legitimate images as outputs. The input and output
of the method in [26] are VQ-images (i.e., a VQ index table),
but Chang and Lin used the SMVQ concept to create three
state codebooks G0, G1, and G2 for each block X to hide
a secret bit if possible. A set of special data structures “hit
maps” is necessary to decode the secret data.
The Chang–Lin method uses the concept of SMVQ, as
shown in Fig. 3. For convenience of description, block “X”
refers to the block that is under encoding. Blocks of the first
column and the first row are not processed, and their encoded
values are their original VQ indices. Then, from left to right
and from top to down, each block X is processed by the
side-match concept. For each block X, three state codebooks
G0, G1, and G2 with the same size are created. An example
of constructing G0, G1, and G2 is shown in Fig. 5, where
the size of the state codebook is 4 and the size of the super
codebook is 16. In the initial stage, the codewords of the super
codebook are clustered. G0 is constructed by the side-match
prediction of X, where the four closest codewords are selected
from the codebook by the distance function defined in (1). For
each codeword cwi of G0, the method identifies the closest
codeword of cwi in the same cluster as the corresponding
codeword of G1.
Joint Neighboring Coding
In 2009, Wang and Lu proposed a paper entitled “A path
optional lossless data-hiding scheme based on VQ joint neighboring
coding” [40]. Each adjacent block of the target block
is given a number, called the position flag bits, as shown in
Fig. 8. The method chooses the adjacent blocks of the target
block on the index table to perform joint neighboring coding
according to their corresponding position flag bits. Then the
secret data is embedded into the cover image using a given
initial key and the secret data content. Two paths are used to
increase the capacity: when Path 1 is selected, 2 bits of the
secret data are hidden in each block on average, and when
Path 2 is selected, 3 bits of the secret data are embedded in
each block. Two enumerating methods for the position flag
bits are shown in Fig. 8(a) and (b). The embedding procedure
uses a pseudorandom number, the M-sequence, to improve
data security. According to the secret bits and the M-sequence,
the method computes the difference value between Ia and the
corresponding index which is generated by secret bits and the
M-sequence.
Comparison and Discussion
Table II shows a brief summary of some representative
papers which were introduced in the previous sections. When
we see deep into each of the methods, we find out that most
researchers aim to design a method that can produce legitimate
VQ-based codes as outputs and maintain reversibility.
However, they usually choose reversibility over the ability of
producing VQ-based stego-images, and modify some part of
the output format that is a little different from the legitimate
VQ-based image format. For example, most part of the output
format of the method in [39] follows the SOC format, and
only one case uses an extended encoding rule.
In the area of image data hiding, people concern about
the quality of the stego-images (if any), the capacity of the
embedded secret data, and the BR of the output that is
delivered in a communication channel. These criteria can be
evaluated by peak signal-to-noise ratio (PSNR) in dB, capacity
in bits, and BR in bits per pixel (b/p), respectively.
Conclusion
Data hiding for digital images is important because it can
protect data and communication against malicious attacks,
536 IEEE SYSTEMS JOURNAL, VOL. 5, NO. 4, DECEMBER 2011
such as information stealing and copyright piracy. A VQ-based
data-hiding method reads a cover image C and a secret data
string S as the input, and creates a stego-image or a code
stream as the output O. VQ-based data-hiding methods usually
provide reversible data hiding, referring to that the output O
can be used to reconstruct the original cover image C and
the secret data string S. This paper presented existing datahiding
methods for VQ-based images, including VQ, SMVQ,
and SOC images.