23-09-2016, 11:54 AM
1455814467-ACABACBasedHEVCVideoSteganographyAlgorithmwithoutBitrateIncrease.pdf (Size: 2.04 MB / Downloads: 4)
Abstract
Video steganography is an important tool for secure communication, which should be closely related
with video coding standard. By investigating the details of context-adaptive binary arithmetic coding
(CABAC) in the entropy coding of the latest high efficiency video coding (HEVC) standard, a novel
concept of “constant bitrate information bit” (CBIB) is presented. For the motion vector difference
(MVD) in HEVC, there are a large amounts of constant-bitrate information bits which can exploited
for video steganography. By designing a codeword reservation and substitution rule for the encoding of
MVDs, a CABAC-based video steganography algorithm is proposed for HEVC video. Video steganography
is conducted in the stage of entropy coding, which is the last step of video compression. It is efficient
because no complex operations such as rate distortion optimization (RDO) or full decoding are involved.
Moreover, since it is based on bit substitution with constant bitrate, there is not any bitrate increase
caused by video steganography. Experimental results on several test sequences prove the effectiveness of
the proposed video steganography approach specifically designed for HEVC.
Introduction
Steganography is the art to hide secret information behind text, audio, and video. There are massive
information amount of digital video signals, which makes video steganography an important
tool for secure communication and copyright protection [1]. Meanwhile, efficient video coding is
needed to make video communication feasible. Thus, most digital video contents are encoded
and stored in the form of compressed bit-stream. In general, video steganography algorithms are
usually combined with video encoder. According to the embedding stage of compression or the
location of syntax elements, video steganography algorithms can be divided into three categories.
That is, secret information in embedded in the stages of transform coding, inter/intra mode
prediction and entropy coding. The syntax elements for data hiding are discrete sine transform
(DCT) coefficients [2, 3], prediction modes [4, 5], and motion vectors [6], respectively.
Up to present, most video steganography are designed for MPEG-x and H.26x. The most
representative works are summarized as follows. To improve the visual effect of video, a data
hiding algorithm is proposed for H.264/AVC video streams without intra- frame distortion drift
[2]. Several paired-coefficients of a 4x4 DCT block are exploited to accumulate the embedding
induced distortion. The directions of intra-frame prediction are utilized to avert the distortion
drift. We also present an information hiding algorithm based on intra-prediction modes and
matrix coding for H.264/AVC video stream [5]. Intra-4x4 coded blocks (I4-blocks) are divided
into groups and two watermark bits are mapped to every three I4-blocks by matrix coding to map
between watermark bit and intra-prediction modes. It can guarantee a high PSNR and slight
bitrate increase after data hiding. To further enhance the security, embedding position template
is utilized to select candidate I4-blocks.
In recent years, HEVC has been developed as a new video coding standard mainly focusing
on the coding of Ultra High Definition (UHD) videos as the high resolution and high quality
videos are getting more popular. As a successor to H.264/AVC standard, HEVC can achieve outstanding
compression performance for HD and UHD video [7]. HEVC is expected to become the
mainstream video coding standard in the next 5 to 10 years. Apparently, video steganography for
HEVC video is worthy of investigation because of both theoretical value and practical application
potential. Since HEVC is a newly developed video coding standard, the video steganography
algorithms specifically designed for digital video encoded by HEVC is still in scarcity. To the best
of our knowledge, there are only two approaches in the literature. An error propagation free data
hiding algorithm is proposed for HEVC intra-coded frames [8]. Since HEVC framework adopts
both DCT and discrete sine transform (DST) for transform coding, it is actually extending the
idea of DCT coefficients-based data hiding approach for H.264/AVC [2] to DCT/DST coefficientsbased
data hiding approach for HEVC. In addition, a large-capacity information hiding method
is proposed for HEVC Video [9], it follows the idea of modulating intra-prediction mode for data
hiding [5]. The mapping rule between prediction modes and secret information is established by
utilizing the probability distribution of the statistically optimal and suboptimal prediction modes.
In this paper, a novel video steganography algorithm without bit-rate increase is presented for
HEVC video. Secret information in embedded in the stage of entropy coding by exploiting the
new features of CABAC for HEVC. The contributions of the proposed approach are two-folds.
First, by analyzing the syntax of CABAC, a novel concept of “constant bitrate information bit”
(CBIB) is introduced. CBIB implies that the mutual substitution between 0 and 1 remains the
bitstream syntax compliance but does not change the bitrate after bit substation. Second, a
replacement rule is specified for the MVD codeword to achieve data hiding. Since data hiding
is performed in the stage of entropy coding, which is the last step of video compression, the
proposed approach does not need any computationally intensive processing such as RDO and
full-decoding of input video stream and can achieve real-time performance. Moreover, since data
hiding is realized by utilizing the CBIBs of CABAC syntax elements, it does not have any bitrate
increase. It is particularly suitable for broadcast monitoring because no re-packetizing is required
for video stream.
The rest paper is organized as follows: Section 2 discusses in detail the CABAC entropy coding
for HEVC, and the concept of CBIB is introduced; Section 3 present the proposed video steganography
approach for HEVC video using the CBIBs; Experimental results and performance analysis
are reported in Section 4, and we conclude this paper in Section 5.
CABAC Entropy Coding of HEVC
Entropy coding is a lossless compression step at the last stage of video encoding (and the first
stage of video decoding). By previous steps such as intra/inter prediction, transform coding and
quantization, video data has been reduced to a series of syntax elements. These syntax elements
include prediction modes, prediction residue and motion vectors. They are needed to be further
compressed by entropy coding to be used at the decoder for frame reconstruction. CABAC is an
entropy coding method firstly presented in H.264/AVC to improve coding efficiency. Actually,
it maps the symbols (i.e., syntax elements) to codewords with a non-integer number of bits.
For the latest video coding standard HEVC, CABAC is also adopted as the base entropy coding.
Specifically, CABAC involves three main functions: binarization, context modeling and arithmetic
coding. Binarization maps the syntax elements to binarize symbols (bins). Context modeling
estimates the probability of the bins. Finally, arithmetic coding is used to compress the bins into
bits based on the estimated probability (context coded) or equal probability of 0.5 (bypass coded).
Fig. 1 shows the schematic diagram of CABAC in HEVC, which highlights the main functions
and the data dependencies. The updated context is fed back for recursive interval division and
accurate probability estimation, and there is no influence on bypass encoding. Therefore, any
changes will make the coding stream structure change, and the encoded video cannot playback
correctly [10]. From the above analysis, we can conclude that binarization is the only outlet
available for video steganography during the CABAC encoding process.
Though both H.264/AVC and HEVC adopt CABAC for entropy coding, there are some differences
between them. Compared with H.264/AVC, the maximum number of context-coded bins
is reduced by 8 [11] in HEVC. The reduction of bins mainly comes from the binarization process.
For HEVC, four different binarization processes are used, including unary (U), truncated unary
(TU), fixed length (FL), k-th-order Exp-Golomb (EGK). Let N be a symbol to be encoded which
has 8 possible values from 0 to 7. An example is provided in Table 1 to show the difference
among these four binarization schemes. Unary coding (U) is quite simple because it uses the
first N bits “1” as the first bin and the last bin “0” as the terminator. Truncated unary coding
(TU) is similar with U. The only difference is that when N + 1 = cM ax, all bins are “1”. The
decoder searches for a bin with a series of “0” up to cMax to determine when the syntax element
is completed. For FL, it uses a fixed length of bins. That is, ceil (log2 (cM ax + 1)). For EGK,
it is a little complex because it is made up of three parts: