Reliable information hiding based on support vector machine.docx

1、Reliable information hiding based on support vector machineReliable Information Hiding Based on Support Vector MachineYong-Gang Fu1,3, Rui-Min Shen1, Li-Ping Shen1 and Xu-Sheng Lei21Dept. of Computer Science and Engineering, Shanghai Jiaotong Univ., Shanghai, China, 2000302Dept. of Automation, Shang

2、hai Jiaotong Univ., Shanghai, China, 2000303Dept. of Software, Xiamen Univ., Xiamen, Chian, 361000E-mails: fyg, rmshen,lpshen, xushengleiAbstract: In this paper, a reliable information hiding scheme based on support vector machine and error correcting codes is proposed. To extract the hidden informa

3、tion bits from a possibly tampered watermarked image with a lower error probability, information hiding is modeled as a digital communication problem, and both the good generalization ability of support vector machine and the error correction code BCH are applied. Due to the good learning ability of

4、 support vector machine, it can learn the relationship between the hidden information and corresponding watermarked image; when the watermarked image is attacked by some intentional or unintentional attacks, the trained support vector machine can recover the right hidden information bits. The reliab

5、ility of the proposed scheme has been tested under different attacks. The experimental results show that the embedded information bits are perceptually transparent and can successfully resist common image processing, jitter attack, and geometrical distortions. When the host image is heavily distorte

6、d, the hidden information can also be extracted recognizably, while most of existing methods are defeated. We expect this approach provide an alternative way for reliable information hiding by applying machine learning technologies.Key words: Information hiding; Support vector machine; Digital water

7、marking; BCH coding1. IntroductionDigital properties are readily reproduced and redistributed over the Internet and other medias. However these attractive properties lead to problems enforcing copyright protection issues. As a result, the contributor and distributor of the digital properties are hes

8、itant to provide the access to their intellectual properties. It is realized that conventional cryptographic means are not sufficient since the data is without any protection as soon as it is used, e.g., decrypted and displayed in the case of image or video data. A potential approach to solve this p

9、roblem is information hiding or digital watermarking (Swanson et al., 1998). Information hiding is the imperceptible embedding of information bits (signature) into multimedia data, where the information remains detectable as long the quality of the content itself is not rendered useless. As a branch

10、 of information hiding, it is commonly assumed that digital watermarking is only one of several measures that have to be combined to build a good copy protection mechanism (Furon and Duhamel, 2000). A significant merit of digital watermarking over traditional protection methods is to provide a seaml

11、ess interface, so that users are still able to utilize the protected multimedia transparently. An information hiding scheme should at least meet the following requirements: (1) Perceptual invisible (or transparent). (2) Difficult to remove without seriously affecting the image quality. (3) Robust re

12、sistance to image processing, and attacks. Developing an algorithm capable of producing signature that fulfills all these requirements is not an easy task. On one hand, the information hiding process should not introduce any perceptible artifacts into the host image. On the other hand, for high robu

13、stness it is desirable that the mark amplitude is as high as possible. Therefore, the designation of information hiding method always involves a tradeoff between imperceptibility (or transparency) and robustness. A variety of watermarking or information hiding schemes have been reported recently in

14、the literature, and some nice reviews can be found in (Fabien et al., 1999). However, the research on copyright protection of images is still in its early stage and none of the existing methods is totally effective against malicious attacks. There are a variety of schemes for hiding information into

15、 the original image. Typical schemes for the information hiding in images can be broadly classified into two categories: (i) spatial domain methods which embed the data by directly modifying the pixel values of the original image (Nikolaidis and Pitas, 1998); (ii) transform domain methods which embe

16、d the data by modulating the coefficients of properly chosen transform domain such as DCT (Cox et al., 1997; Barni et al., 2000), DFT (Barni et al., 2000), and DWT (Xia et al., 1998). Many of the spatial domain techniques provide simple and effective schemes for embedding an invisible watermark into

17、 an image but are not robust to common attacks. Information hiding techniques can be alternatively split into two distinct categories depending on whether the original image is necessary for the watermark extraction or not. Although the existence of the original image facilitates watermark extractio

18、n (Cox et al., 1997; Swanson et al.1996; Podilchuk and Zeng, 1998) to great extent, such a requirement raises two problems: (i) owner of the original image is compelled insecurely to share his works with anyone who wants to check the existence of the signature (Barni et al., 1998), (ii) on the other

19、 hand, the searching within the database for the original image that corresponds to a given watermarked image would be very time consuming. Thus, methods capable of revealing the information bits presence without comparing the watermarked and original images would be preferable. In order to design r

20、obust information hiding scheme, Cox et al. considered watermarking as a problem of communication with side information (Cox et al., 1999). Also, some watermarking algorithm in literature applied error correcting coding(ECC) to improve the bit error rate(BER) performance, such as Bose-Chaudhuri-Hocq

21、uenghen(BCH) coding (Huang et al.1998; Huang and Yun, 2002), Reed-Solomon(R-S) code (Wu and Hsieh, 2000), and Turbo code (Pereira and Pun, 2000). Recently, efforts are made to use artificial intelligence technique for watermark embedding and extraction. Neural networks are introduced into watermarki

22、ng in (Yu et al., 2001), which makes the watermark extraction more robust against common attacks. Genetic algorithm is proposed for selection of the best embedding positions in block based DCT domain watermarking (Shieh et al., 2004). We have firstly introduced the support vector machine for the wat

23、ermark embedding and extraction in (Fu et al., 2004), in which the watermark is embedded into the host by applying the good learning ability of support vector regression machine, and the watermark extraction is finished by the aids of the well trained support vector machine. We can expect that the c

24、ombination of information hiding and machine learning techniques might be a good solution for reliable information hiding. From the observations above, in this paper we propose a novel blind reliable information hiding and recovering scheme which makes use of support vector machine and BCH coding. T

25、his work can be considered as an extension of some existing research (Kutter et al.,1998; Yu et al.,2001; Fu et al., 2004). In (Kutter et al., 1998), Kutter proposed a spatial domain watermarking scheme for color image. Then Yu (Yu et al., 2001) improved Kutters method by applying neural networks. D

26、ue to the support vector machines good learning ability in training process, it can memorize the relationship between the embedded information and corresponding watermarked image. Applying SVMs good generalization abilities and error correcting ability of BCH coding, hidden information extraction ca

27、n be finished well. Experimental results show good robustness of the proposed scheme against common image processing and attacks. This research is much different from my early work. In this research, the support vector machine is only trained and applied in the information extraction procedure, wher

28、eas, in (Fu et al., 2004), the support vector regression machine is applied both in the watermark embedding and extraction process.The paper is organized as follows: in section 2, basic conceptions for support vector machine are introduced. The embedding and extraction algorithms of our method are d

29、escribed in section 3. In section 4, some experimental results are exhibited. The conclusion is stated in section 5.2. An overview of Support Vector MachineSupport Vector Machine (SVM) is a universal classification algorithm developed by Vapnik and his colleagues (Vapnik, 1995; Vapnik, 1998). In rec

30、ent years, there have been a lot of interests in studying the applications of SVM on function approximation, pattern recognition problems and so on (Campbell, 2002; Christopher, 1998).Given a training data set of m samples, whereis the ith input pattern andis the ith output pattern. The support vect

31、or machine method supposes we have some hyper-planes that separate the positive samples from negative ones. The pointwhich lies on the hyper-plane satisfies whereis normal to the hyper-plane, is the perpendicular distance from the hyper-plane to the origin, and is the Euclidean norm of. For the line

32、arly separable case, the support vector algorithm simply looks for the separating hyper-plane with the largest margin. This can be formulated as following: suppose that all the training data satisfy the following constraints: (1)This can be combined into one set of inequalities: (2)Now consider the points for which the equality in (1) holds. These points lie on the hyper-plane with normaland perpendicular distance from the origin. Similarly, the points for which the equality holds in (1) lie on the hyper-plane, with normal, and perpendicular distance from the origin. Hence the