Elsevier

Pattern Recognition

Volume 37, Issue 3, March 2004, Pages 555-565
Pattern Recognition

Genetic watermarking based on transform-domain techniques

https://doi.org/10.1016/j.patcog.2003.07.003Get rights and content

Abstract

An innovative watermarking scheme based on genetic algorithms (GA) in the transform domain is proposed. It is robust against watermarking attacks, which are commonly employed in the literature. In addition, the watermarked image quality is also considered. In this paper, we employ GA for optimizing both the fundamentally conflicting requirements.

Watermarking with GA is easy for implementation. We also examine the effectiveness of our scheme by checking the fitness function in GA, which includes both factors related to robustness and invisibility. Simulation results also show both the robustness under attacks, and the improvement in watermarked image quality with GA.

Introduction

With the widespread use of Internet and the development in computer industry, the digital media, including images, audios, and video, are easily acquired in our daily life. Digital multimedia contents suffer from infringing upon the copyrights with the digital nature of unlimited duplication, easy modification and quick transfer over the Internet. As a result, data piracy has become a serious issue. Hence, some copyright protection schemes need to be employed to conquer this problem. In this paper, we concentrate our research topic on image watermarking for copyright protection.

Digital watermarking for images is one way to embed the secret information, or the watermark, into the original image itself to protect the ownership of the original sources [1], [2], [3]. On the one hand, the watermarking schemes can be categorized as “visible” and “invisible” watermarking. Comparisons between the two schemes are listed as follows:

  • 1.

    The visible watermarks, for instance, are those company logos on one corner of the TV screen when we watch TV programs. Although the logos, or watermarks, are easily identified, they are usually not robust against image cropping. Therefore, visible watermarks can be easily removed from original images.

  • 2.

    The invisible watermarks are more secure and robust than the visible watermarks. The embedding locations are secret, and only the authorized persons with the secret keys in the watermarking system can extract the secret watermark. The watermarked image should look similar to the original one, and should not cause suspicion by others.

On the other hand, digital watermarking can also be categorized as “robust” and “fragile” watermarking. Robust watermarks are designed to have the ability to detect the watermark after some image processing operations, called attacks. After certain attacks and the watermark extraction process, the extracted watermarks should be highly correlated with the embedded ones. That is, the extracted watermarks need be recognizable in the robust watermarking system. In contrast, for fragile watermarks, they are designed to become invalid after even the slightest modification of the watermarked image. Because the watermarks become undetectable, they do not resist intentional or unintentional attacks. Therefore, fragile watermarks are mainly used for authentication purposes.

There are a variety of schemes for embedding the watermark into the original image [4]. Typical schemes for digital watermarking were based on transform-domain techniques with discrete cosine transform (DCT) [5], [6], [7], discrete wavelet transform (DWT) [8], discrete Fourier transform (DFT) [9], spatial-domain methods [10], [11], and vector quantization (VQ) domain schemes [12]. The above-mentioned schemes employ the embedding of the watermark into some of the selected coefficients in their corresponding domains, which might be fixed in a pre-determined set of coefficients. One major disadvantage for these typical schemes is that during transmission over the Internet or the mobile channels, the watermarked images might be processed, or attacked, in order to remove the existence of the watermark [13], [14]. When the attackers dissolve the relationships between the original multimedia and the pre-determined set for watermark embedding, the watermarking capability for copyright protection no longer exists. Another disadvantage for typical schemes is how to decide and choose the pre-determined set. For watermark embedding in the DCT domain, if we embed the watermark in the higher frequency bands, even though the watermarked image quality is good, it is vulnerable to the low pass filtering (LPF) attack. Thus, embedding into the higher frequency bands coefficients is not robust, although the watermarked image quality is assured. In contrast, if we embed the watermark into the coefficients in the lower frequency bands, it should be robust against common image processing attacks such as the LPF attack. However, embedding in the lower frequency bands will cause the resulting watermarked image quality greatly degrades to compare with the original image. This comes from the fact that the energies of most natural images are concentrated in lower frequency bands, and the human eyes are more sensitive to the noise caused by modifying the lower frequency coefficients. Hence, aside from the two observations above, some researchers claim to embed the watermarks into the “middle-frequency bands” to serve as a trade-off for watermark embedding in the transform domain [5].

Therefore, from the observations and explanations above, we make use of genetic algorithm (GA) [15], [16] to find the optimal frequency bands for watermark embedding into our DCT-based watermarking system, which can simultaneously improve security, robustness, and image quality of the watermarked image. Because the scheme operates in the transform domain, it contains three main parts, including image transformation, watermark embedding, and watermark extraction.

This paper is organized as follows. We describe the fundamental concepts of genetic algorithms in Section 2. Section 3 demonstrates the algorithm for embedding the watermark in the DCT domain with GA. Section 4 depicts the watermark extraction algorithm. Section 5 illustrates the simulation results, and we also show the superiority of our scheme over the results proposed by other researchers in this section. Section 6 briefly discusses with the proposed algorithm and the simulation results. And we conclude this paper in Section 7.

Section snippets

Fundamental concepts of genetic algorithms

Conventional search techniques are often incapable of optimizing non-linear functions with multiple variables. One scheme called the “genetic algorithm” (GA), based on the concept of natural genetics, is a directed random search technique. The exceptional contribution of this method was developed by Holland [15] over the course of 1960s and 1970s, and finally popularized by Goldberg [16].

In the genetic algorithms, the parameters are represented by an encoded binary string, called the

The embedding algorithm

Let the input image be X with size M×N. Our goal is to embed a robust watermark into the DCT frequency bands of X, and have a watermarked reconstruction X after optimization.

Before the embedding procedure, we need to transform the spatial domain pixels into DCT domain frequency bands. We perform the 8×8 block DCT on X first and get the coefficients in the frequency bands, Y,Y=DCT(X)andY=m=1M/8n=1N/8Y(m,n).For one non-overlapping block (m,n) in X, the resulting 64 DCT bands Y(m,n) can be

The extraction algorithm

In extracting the watermarks, the original image X is not required in our algorithm. However, the optimized watermarked image might be subject to some intentional or unintentional attack, and the resulting image after attack is represented by X. We calculate the DCT of the watermarked image after attacking Y, in the attacked X, with the secret key corresponding to the frequency set F, key1. We then reproduce the estimated reference table R from the attacked X by following the operations in

Simulation results

In our simulation, we take the well-known test image, Lena, with size 512×512, as the original source, which is shown in Fig. 5. We have the embedded watermark, rose, with size 128×128, shown in Fig. 6. Hence, the number of bits to be embedded in one 8×8 non-overlapping block is 12825122·64=4. Next, taking the frequency set in Ref. [5] to be the initial set F, i.e., F={Y(m,n)(14),Y(m,n)(15),Y(m,n)(16),Y(m,n)(27)}, for every block in our simulation. After watermark embedding in the DCT domain,

Discussions

The goal of GA is to find an optimized solution under several conflicting requirements. In this paper, we select the two conflicting requirements for typical watermarking systems, namely, the watermarked image quality and the robustness of the watermarking algorithm. The simulation results in Section 5 also prove the effectiveness of our GA-based watermarking algorithm.

We observed the following points in our discussion for this paper:

  • 1.

    In the fitness function, the parameters to be optimized are

Conclusion

A robust algorithm for DCT-based GA-watermarking has been presented in this paper. It is robust because we make use of GA to train the frequency set for embedding the watermark. In addition to the robustness of the proposed algorithm, we also improve the watermarked image quality with the aid of GA.

Simulation results reveal that if we just borrow the concepts of existing algorithms, both the watermarked image quality and the NC values of the extracted watermarks after certain attacks will be

Acknowledgements

This work was supported by National Science Council (Taiwan, ROC) under Grant No. NSC91-E-2219-151-002.

About the Author—CHIN-SHIUH SHIEH received the B.S. degree in Electronic Engineering from National Taiwan Institute of Technology, Taiwan, in 1989, and the M.S. degree in Electrical Engineering from National Taiwan University, Taiwan, in 1991. His current research interests are in self-learning fuzzy systems using evolutionary techniques, computer networking, and various issues in vector quantization.

References (21)

  • N. Nikolaidis et al.

    Robust image watermarking in the spatial domain

    Signal Process.

    (1998)
  • F.A.P. Petitcolas et al.

    Information hiding—a survey

    Proc. IEEE

    (1999)
  • F. Hartung et al.

    Multimedia watermarking techniques

    Proc. IEEE

    (1999)
  • D. Artz

    Digital steganographyhiding data within data

    IEEE Internet Comput.

    (2001)
  • C.I. Podilchuk et al.

    Digital watermarkingalgorithms and applications

    IEEE Signal Process. Mag.

    (2001)
  • C.T. Hsu et al.

    Hidden digital watermarks in images

    IEEE Trans. Image Process.

    (1999)
  • J.R. Hernandez et al.

    DCT-domain watermarking techniques for still imagesdetector performance analysis and a new structure

    IEEE Trans. Image Process.

    (2000)
  • W.C. Chu

    DCT-based image watermarking using subsampling

    IEEE Trans. Multimedia

    (2003)
  • M. Barni et al.

    Improved wavelet-based watermarking through pixel-wise masking

    Image Process. IEEE Trans. Image Process.

    (2001)
  • J.J.K. O'Ruanaidh, W.J. Dowling, F.M. Boland, Phase watermarking of digital image, Proceedings of the IEEE...
There are more references available in the full text version of this article.

Cited by (291)

  • ELM-Kernel and reduced kernel ELM based watermarking scheme

    2019, Journal of Information Security and Applications
  • A Novel fuzzy frame selection based watermarking scheme for MPEG-4 videos using Bi-directional extreme learning machine

    2019, Applied Soft Computing Journal
    Citation Excerpt :

    Another aspect is to identify and use optimum pixel locations for optimized embedding of the watermark in grayscale and color images in the transform domain. This is usually carried out by using Genetic Algorithm (GA) [31], Particle Swarm Optimization (PSO) [32], Firefly Algorithm (FA) [20], GA-BPN [33] and Cuckoo Search Techniques [34]. To accomplish the real-time criteria, several researchers have implemented the Extreme Learning Machine (ELM) and its different variants such as Online-Sequential [35,36], Bi-directional ELM (B-ELM) [19] for image watermarking.

  • Recent Information Hiding Techniques in Digital Systems: A Review

    2024, Journal of Information Hiding and Multimedia Signal Processing
View all citing articles on Scopus

About the Author—CHIN-SHIUH SHIEH received the B.S. degree in Electronic Engineering from National Taiwan Institute of Technology, Taiwan, in 1989, and the M.S. degree in Electrical Engineering from National Taiwan University, Taiwan, in 1991. His current research interests are in self-learning fuzzy systems using evolutionary techniques, computer networking, and various issues in vector quantization.

About the Author—HSIANG-CHEH HUANG received the B.S., M.S. and Ph.D. degrees in Electronics Engineering from National Chiao Tung University, Taiwan, ROC, in 1995, 1997, and 2001, respectively. Currently, he is a post-doctor researcher in the Department of Electronics Engineering, National Chiao Tung University, Taiwan, ROC. His current research interests include pattern recognition and image processing.

About the Author—FENG-HSING WANG received the B.S. degree in Electronic Engineering from National Kaohsiung University of Applied Sciences, Taiwan, ROC. Currently, he is a Ph.D. candidate in the School of Electrical and Information Engineering, University of South Australia, Adelaide, Australia. His current research interests include computer architecture and image processing.

About the Author—JENG-SHYANG PAN received the B.S. degree in Electronic Engineering from the National Taiwan Institute of Technology, Taiwan, in 1986, the M.S. degree in Communication Engineering from National Chiao Tung University, Taiwan, ROC in 1988, and the Ph.D. degree in Electrical Engineering from the University of Edinburgh, UK, in 1996. Currently, he is a Professor in the Department of Electronic Engineering, National Kaohsiung University of Applied Sciences, Taiwan, ROC. He is also an advisor of postgraduate students both in the Department of Electrical and Information Engineering, University of South Australia and the Department of Automatic Test and Control, Harbin Institute of Technology. He has published more than 35 international journal papers and 70 conference papers. His current research interests include pattern recognition, information security and data mining.

View full text