17-05-2013, 12:24 PM
Robust Reversible Watermarking via Clustering and Enhanced Pixel-Wise Masking
Robust Reversible.pdf (Size: 3.2 MB / Downloads: 40)
Abstract
Robust reversible watermarking (RRW) methods
are popular in multimedia for protecting copyright, while
preserving intactness of host images and providing robustness
against unintentional attacks. However, conventional RRW
methods are not readily applicable in practice. That is mainly
because: 1) they fail to offer satisfactory reversibility on
large-scale image datasets; 2) they have limited robustness in
extracting watermarks from the watermarked images destroyed
by different unintentional attacks; and 3) some of them suffer
from extremely poor invisibility for watermarked images. Therefore,
it is necessary to have a framework to address these three
problems, and further improve its performance. This paper
presents a novel pragmatic framework, wavelet-domain statistical
quantity histogram shifting and clustering (WSQH-SC).
Compared with conventional methods, WSQH-SC ingeniously
constructs new watermark embedding and extraction procedures
by histogram shifting and clustering, which are important
for improving robustness and reducing run-time complexity.
Additionally, WSQH-SC includes the property-inspired pixel
adjustment to effectively handle overflow and underflow of
pixels. This results in satisfactory reversibility and invisibility.
Furthermore, to increase its practical applicability, WSQH-SC
designs an enhanced pixel-wise masking to balance robustness
and invisibility. We perform extensive experiments over
natural, medical, and synthetic aperture radar images to
show the effectiveness of WSQH-SC by comparing with the
histogram rotation-based and histogram distribution constrained
methods.
INTRODUCTION
REVERSIBLE WATERMARKING (RW) methods [1] are
used to embed watermarks [2], e.g., secret information
[3], into digital media while preserving high intactness
and good fidelity of host media. It plays an important
role in protecting copyright and content of digital media
for sensitive applications, e.g., medical and military images.
Although researchers proposed some RW methods for various
media, e.g., images [4], [5], audios [6], videos [7], and 3-D
meshes [8], they assume the transmission channel is lossless.
The robust RW (RRW) is thus a challenging task. For RRW,
the essential objective is to accomplish watermark embedding
and extraction in both lossless and lossy environment. As a
result, RRW is required to not only recover host images and
watermarks without distortion for the lossless channel, but also
resist unintentional attacks and extract as many watermarks as
possible for the noised channel [9]. Recently, a dozen of RRW
methods for digital images have been proposed [10]–[13],
which can be classified into two groups [9]: histogram rotation
(HR)-based methods and histogram distribution constrained
(HDC) methods.
RELATED WORKS
In this section, we briefly introduce the GSQH driven
method [14] and discuss its useful inspirations to our novel
framework. Thereafter, a popular pixel-wise masking (PWM)
model is presented to lay the groundwork for the proposed
EPWM.
GSQH Driven Method
The histogram plays an important role in many practical
models and applications, e.g., histogram of oriented
gradient features [15], bag-of-words [16], [17], and digital
watermarking [18]–[20]. For RW methods, SQH has recently
received considerable attention due to stability and simplicity,
e.g., arithmetic average of difference (AAD) histogram [21],
difference histogram [22], and prediction error histogram [23].
In particular, we proposed a GSQH driven method [14], which
embeds and extracts watermarks by SQH shifting. The
following is a brief review of this method.
PROPOSED FRAMEWORK
In this section, we introduce a new RRW framework, i.e.,
WSQH-SC, which accomplishes the robust lossless embedding
task by incorporating the merits of SQH shifting, k-means
clustering and EPWM. WSQH-SC comprises two processes:
watermark embedding and extraction. In view of their similarity,
Fig. 1 only shows the diagram of the embedding process
in which the three modules are termed: 1) PIPA; 2) SQH
construction; and 3) EPWM-based embedding, and they are
detailed in the following three subsections. To be specific,
WSQH-SC first investigates the wavelet sub-band properties
in depth and exploits PIPA to preprocess the host image,
which is of great importance to avoid both overflow and
underflow of pixels during the embedding process. Afterward,
the host image is decomposed by the 5/3 integer wavelet
transform (IWT) [30] and the blocks of interest in the sub-band
cHL
0 are selected to generate the SQH with the help of the
threshold constraint. Finally, watermarks can be embedded into
the selected blocks by histogram shifting, wherein EPWM is
designed to adaptively control watermark strength. After the
IWT reconstruction, the watermarked image is obtained.
SQH Construction
In this subsection, we consider the SQH construction task
with a threshold constraint. Inspired by characteristics of
the wavelet coefficients [13], we focus on the mean of
wavelet coefficients (MWC) histogram by taking the following
two properties into account: 1) it is designed in high-pass
sub-bands of wavelet decomposition, to which HVS is less
sensitive, leading to high invisibility of watermarked images
and 2) it has almost a zero-mean and Laplacian-like distribution
based on the experimental study of wavelet high-pass
sub-bands from 300 test images illustrated in Section V, which
is stable for different images. In particular, an MWC histogram
is generated based on the following procedure.
Extraction Based on k-Means Clustering
If watermarked images are transmitted through an ideal
channel, we can directly adopt the inverse operation of (19)
to recover host images and watermarks. However, in the real
environment, degradation may be imposed on watermarked
images due to unintentional attacks, e.g., lossy compression
and random noise. Therefore, it is essential to find an effective
watermark extraction algorithm so that it can resist unintentional
attacks in the lossy environment.
CONCLUSION
In this paper, we have developed a novel yet pragmatic
framework for RRW. It includes carefully designed PIPA,
SQH shifting and clustering, and EPWM, each of which
handles a specific problem in RRW. PIPA preprocesses host
images by adjusting the pixels into a reliable range for satisfactory
reversibility. SQH shifting and clustering constructs
new watermark embedding and extraction processes for good
robustness and low run-time complexity. EPWM precisely
estimates the local sensitivity of HVS and adaptively optimizes
the watermark strength for a trade-off between robustness and
invisibility. In contrast to representative methods, thorough
experimental results on natural, medical and SAR images
demonstrate that the proposed framework: 1) obtains comprehensive
performance in terms of reversibility, robustness,
invisibility, capacity and run-time complexity; 2) is widely
applicable to different kinds of images; and 3) is readily
applicable in practice.