Research On Collusion Attack Optimization Algorithm For Digital Fingerprinting

With the rapid development of multimedia and Internet, especially the emergence of various of social network services, the multimedia data has been increasing explosively. This brings people a lot of convenience to share and get multimedia content from Internet. However, these benefits also bring ease to unauthorized use of multimedia content, such as illegal duplication, processing and redistribution, which leads to the copyright infringement and privacy leakage. Thus, the most essential problem is how to protect the copyright of multimedia data and how to prevent the illegal usage and dispersion.Digital fingerprinting is considered as an efficient and most potential method to solve the aforementioned problem. Unique marks, known as fingerprints, are embedded in the content which is used to identify adversaries who leak the copies of the same content. How-ever, collusion attack is known to be a cost-effective attack and poses a great threat to the digital fingerprinting systems. In collusion attack, a group of users combines multiple copies of the same multimedia content to generate a new version. With enough number of col-lected copies, the adversaries arc able to attenuate or remove the fingerprints, which makes the detector unable to trace any of the real colluders involved. Therefore, understanding the weaknesses and the limitations of existing fingerprinting schemes and designing effective collusion attacks play an important role in the development of digital fingerprinting.At first, we focus on studying the typical digital fingerprinting schemes and commonly used collusion attacks, and construct an general evaluation framework based upon the se-curity analysis of digital fingerprinting. Under this framework, we can evaluate the perfor-mance of typical digital fingerprinting schemes when suffering frequently used linear and non-linear collusion attacks.Then, we propose there novel collusion attack optimization algorithms from three dif-ferent perspectives. At first, combining with the geometric representation of Voronoi dia-gram, we propose an Iterative Optimization Collusion Attack (IOCA). By using the iterative optimization method, we remove the fingerprint which can be detected as containing the col- luders in the Voronoi diagram, after processing of several iterations, to a signal which can be detected as not. After that, we evaluate the performance of the proposed collusion attack strategy in defeating four typical fingerprinting schemes under a well-constructed evaluation framework.Although the IOCA is more effective in defeating the fingerprinting system than the traditional collusion attacks, it also introduces larger noticeable distortion. To improve the perceptual quality of the attacked signal, we propose a self-adaptive noising optimization (SANO) collusion attack based on the fact that the probability of the detector to be easily distorted by the noise. We at first analyze the influence of the Gaussian noise on both the detection probability of the correlation detector and the perceptual quality of the attacked content. Then, based upon the principle of the closed-loop feedback control theory, an adap-tive dynamic iterative approach is applied to add the noise to the forgery until obtaining a copy where the fingerprint is not presented or making the detector difficult to detect any of the pirates. The experimental results show that the proposed algorithm can efficiently inter-rupt the fingerprinting system with less than three independent copies of the same content.From the perspective of the fingerprint embedding, we know that the fingerprint is in reality a noise signal added to the original content. Therefore, we propose a support vector machine (SVM)-based anti-forensic method by employing denoising approach. We at first train the SVM-based classifier using fingerprinted and un-fingerprinted copies, and obtain the optimal classification parameters. After that, we select the best basis through wavelet packet decomposition for thresholding the fingerprints from the forgery.Theory analysis and experimental results illustrate that the proposed collusion attack optimization algorithms are more effective than usually used linear and nonlinear collusion attacks and the gradient attack. However, we assume that the fingerprinting extracting algorithm is known to the colluders in the proposed algorithms. It is worthwhile to explore an effective collusion attack strategy under the assumption that the fingerprinting extracting algorithm is unknown.