17-03-2014, 06:40 PM
Abstract—In this paper, a novel blind watermarking scheme
for still image based on the two-dimensional discrete wavelet
transform (DWT) is proposed. In the scheme, digital
watermarking with a pseudo-random sequence is derived
from a full Logistic map system, and the robustness is
achieved by embedding every bit of watermark sequence
several times in many different locations in approximate
sub-band of one-level Haar wavelet decomposition. The
strength of watermarking is adjusted by an auto-adaptive
way for fidelity constraints. During detection process, by use
of the statistic characteristic of sub-band LL coefficients and
the maximum subjection principle, the watermarking is
extracted reliably without recurring to original image.
Simulations demonstrate that the proposed method can
resist many kinds of signal processing attacks such that
JPEG compression, additive noise, linear filtering, format
conversion, cropping and slight geometric distortion, and so
the proposed algorithm has high robustness.
Keywords-blind watermarking; robustness; statistic
characteristic; DWT
I. INTRODUCTION
With the rapid development of digital multimedia, the
talented works are created and distributed by various
digital methods. Protecting the copyright of the digital
media has become an important topic due to digital media
can be copied and modified easily. As an effective
technique for protecting the copyright, digital
watermarking technique has received much attention in the
past decades, and various algorithms have been proposed,
such as robust watermarking algorithm [1-2], fragile
watermarking algorithm [3-4] and reversible watermarking,
which is suitable for important digital media [5-6].
As an important type of watermarking technique,
robust watermarking demands that when digital media
suffers from various attacks such as JPEG encoding, linear
filter, additive noise, quantization, cropping etc.,
embedded watermark can still be extracted and identified
[7]. To meet the requirements of robustness, watermark is
often embedded into the transform domain rather than
spatial domain. In the corresponding research aspects, Cox
et al. discussed several approaches to make watermarks
robust [8]. It is emphasized that watermarks must be
embedded in unsusceptible spectrum domain coefficients
incorporating redundant embedding in order to preventing
from failed detection caused by intended processing. As a
guideline, this law has been leading to many useful robust
watermarking algorithms released for a long time. Cox et
al. also proposed a robust, spread-spectrum method using a
full-frame discrete cosine transform (DCT), the watermark
consists of a random sequence chosen from a normal
distribution and is embedded by adding it to some
purposely selected AC coefficients [9]. Xia et al.
introduced a watermarking scheme based on the Discrete
Wavelet Transform, the watermark, modeled as Gaussian
noise, is added to the middle and high frequency bands of
the image, the decoding process involved taking the DWT
of a potentially marked image [10]. Ruanaidh and Pun
proposed to combine Fourier transform with Log Polar
Mapping (LPM), this associated transform provides
robustness against common affine distortion but not
implement a good performance in resisting compression
and blurring manipulation [11]. Furthermore, LPM and
inverse LPM are lossy operations that may cause
degradations in the watermark image and exhaust precious
computational cost.
In this paper, a new blind robust watermarking
algorithm is developed by employing a statistic
characteristic of low frequency sub-band coefficients of
one-level Haar wavelet transform, which is demonstrated
to be more appropriate orthogonal wavelet base for digital
image wavelet decomposition [12]. Large amounts of tests
over many approximate sub-band coefficients of nature
images testify that the LL sub-image is a totally smooth
region (i.e. Unless a few coefficients representing edge or
texture, the coefficient in any position is very close to its
four cross orientation neighbor coefficients on value).Thus,
in general, the difference of center coefficient and the
mean of four cross neighbor coefficients is always less
than some threshold. This is the basis of watermarking
embedding and extracting. In addition, wavelet plays an
important role in the upcoming image compression
standard such as JPEG2000 and lower computational cost
than DFT or DCT.
for still image based on the two-dimensional discrete wavelet
transform (DWT) is proposed. In the scheme, digital
watermarking with a pseudo-random sequence is derived
from a full Logistic map system, and the robustness is
achieved by embedding every bit of watermark sequence
several times in many different locations in approximate
sub-band of one-level Haar wavelet decomposition. The
strength of watermarking is adjusted by an auto-adaptive
way for fidelity constraints. During detection process, by use
of the statistic characteristic of sub-band LL coefficients and
the maximum subjection principle, the watermarking is
extracted reliably without recurring to original image.
Simulations demonstrate that the proposed method can
resist many kinds of signal processing attacks such that
JPEG compression, additive noise, linear filtering, format
conversion, cropping and slight geometric distortion, and so
the proposed algorithm has high robustness.
Keywords-blind watermarking; robustness; statistic
characteristic; DWT
I. INTRODUCTION
With the rapid development of digital multimedia, the
talented works are created and distributed by various
digital methods. Protecting the copyright of the digital
media has become an important topic due to digital media
can be copied and modified easily. As an effective
technique for protecting the copyright, digital
watermarking technique has received much attention in the
past decades, and various algorithms have been proposed,
such as robust watermarking algorithm [1-2], fragile
watermarking algorithm [3-4] and reversible watermarking,
which is suitable for important digital media [5-6].
As an important type of watermarking technique,
robust watermarking demands that when digital media
suffers from various attacks such as JPEG encoding, linear
filter, additive noise, quantization, cropping etc.,
embedded watermark can still be extracted and identified
[7]. To meet the requirements of robustness, watermark is
often embedded into the transform domain rather than
spatial domain. In the corresponding research aspects, Cox
et al. discussed several approaches to make watermarks
robust [8]. It is emphasized that watermarks must be
embedded in unsusceptible spectrum domain coefficients
incorporating redundant embedding in order to preventing
from failed detection caused by intended processing. As a
guideline, this law has been leading to many useful robust
watermarking algorithms released for a long time. Cox et
al. also proposed a robust, spread-spectrum method using a
full-frame discrete cosine transform (DCT), the watermark
consists of a random sequence chosen from a normal
distribution and is embedded by adding it to some
purposely selected AC coefficients [9]. Xia et al.
introduced a watermarking scheme based on the Discrete
Wavelet Transform, the watermark, modeled as Gaussian
noise, is added to the middle and high frequency bands of
the image, the decoding process involved taking the DWT
of a potentially marked image [10]. Ruanaidh and Pun
proposed to combine Fourier transform with Log Polar
Mapping (LPM), this associated transform provides
robustness against common affine distortion but not
implement a good performance in resisting compression
and blurring manipulation [11]. Furthermore, LPM and
inverse LPM are lossy operations that may cause
degradations in the watermark image and exhaust precious
computational cost.
In this paper, a new blind robust watermarking
algorithm is developed by employing a statistic
characteristic of low frequency sub-band coefficients of
one-level Haar wavelet transform, which is demonstrated
to be more appropriate orthogonal wavelet base for digital
image wavelet decomposition [12]. Large amounts of tests
over many approximate sub-band coefficients of nature
images testify that the LL sub-image is a totally smooth
region (i.e. Unless a few coefficients representing edge or
texture, the coefficient in any position is very close to its
four cross orientation neighbor coefficients on value).Thus,
in general, the difference of center coefficient and the
mean of four cross neighbor coefficients is always less
than some threshold. This is the basis of watermarking
embedding and extracting. In addition, wavelet plays an
important role in the upcoming image compression
standard such as JPEG2000 and lower computational cost
than DFT or DCT.