Soft Comput (2009) 13:355–360
DOI 10.1007/s00500-008-0331-y
FOCUS
A new watermarking approach based on probabilistic
neural network in wavelet domain
Xian-Bin Wen · Hua Zhang · Xue-Quan Xu ·
Jin-Juan Quan
Published online: 7 June 2008
© Springer-Verlag 2008
Abstract A novel scheme of digital image watermarking
based on the combination of dual-tree wavelet transform
(DTCWT) and probabilistic neural network is proposed in
this paper. Firstly, the original image is decomposed by
DTCWT, and then the watermark bits are added to the selec-
ted coefficients blocks. Because of the learning and adaptive
capabilities of neural networks, the trained neural networks
can recover the watermark from the watermarked images.
Experimental results show that the proposed scheme has
good performance against several attacks.
Keywords Watermarking · Wavelet ·
Probabilistic neural network
1 Introduction
With the rapid development of Internet, digital properties
are readily reproduced and distributed with ease. However,
these attractive properties lead to problems enforcing copy-
right protection while exchanging multimedia data over the
Internet. There are already great deals of discussions about
how to protect the rights of the creator/owner. It is real-
ized that conventional cryptographic means are not suffi-
cient since the data is without any protection as soon as it
is used, e.g., decrypted and displayed in the case of image
X.-B. Wen (
B
) · H. Zhang · X.-Q. Xu · J.-J. Quan
School of Computer Science and Technology,
Tianjin University of Technology,
300191 Tianjin, People’s Republic of China
e-mail: xbwen@tjut.edu.cn
X.-B. Wen · H. Zhang · X.-Q. Xu · J.-J. Quan
Tianjin Key Laboratory of Intelligence Computing
and Novel Software Technology, 300191
Tianjin, People’s Republic of China
or video. A potential approach to solve this problem is the
digital watermarking technique. A significant merit of digital
watermarking is that multimedia data can still be utilized by
users although they are embedded with an invisible digital
watermark. These watermarks cannot be removed by unau-
thorized persons and they can be extracted by legal author.
A watermarking scheme should at least meet the follow-
ing requirements: (1) Perceptual invisible (or transparency).
(2) Difficult to remove without seriously affecting the image
quality. (3) Robust against image processing, and attacks.
A variety of watermarking or information hiding schemes
have been reported recently in the literature and some nice
surveys can be found in reference (Mohanty 1999). How-
ever, research on copyright protection of images is still in its
early stage and none of the existing methods is totally effec-
tive against malicious attacks. Watermarking techniques for
digital images have been reported in the past. Watermarking
techniques can be broadly classified into two categories: spa-
tial domain methods (Yu et al. 2001) and transform domain
methods (Chen et al. 2005; Khelifi et al. 2005; Nafornita
2005; Temi et al. 2005). Spatial domain methods are sim-
ple and fast, but are not robust against attacks. Embedding
the watermark into the transform domain can increase the
security, imperceptibility and robustness of watermark, and
is widely adopted in many digital watermark methods.
Recently, efforts are made to take advantage of artifi-
cial intelligence techniques for watermark embedding and
extraction (Pan et al. 2007, 2004; Chang et al. 2007; Shieh
et al. 2004; Fu et al. 2004; Wang et al. 2006; Zhang and
Zhang 2005; Zhang et al. 2002; Xu et al. 2007; Tseng et al.
2004). Neural networks are introduced into watermarking
(Wang et al. 2006; Zhang and Zhang 2005; Zhang et al. 2002;
Xu et al. 2007; Tseng et al. 2004),whichmakethewater-
mark detection more robust against common attacks. Genetic
algorithm is proposed for selection of the best embedding
123