Novel Algorithms for Sensitivity Analysis

Sensitivity analysis attacks constitute a known family of watermark “removal” attacks exploiting a vulnerability in some watermarking protocols: the attacker’s unlimited access to the watermark detector. In this work, novel attacks on additive spread spectrum schemes are designed. We first examine a well-known algorithm by Kalker et al., and prove its convergence using the law of large numbers; this analysis provides more insight into the problem. Next, two new algorithms are presented and compared to existing ones. The new algorithms successfully “remove” the watermark for a wide range of detection methods. These include the simple correlation detectors and normalized correlation detectors, as well as other, more complicated detectors like the maximum likelihood detector corresponding to generalized Gaussian hosts. Moreover, the suggested algorithms are noniterative and require at most 2n + 1 detection operations in order to estimate the watermark, where n is the dimension of the signal. Briefly, the new approach is to directly estimate the watermark by exploiting the geometry of the detection boundary and the information leaked by the detector. These facts illustrate the high susceptibility of one of the most used embedding schemes (additive spread spectrum) and emphasize the vital need for more secure watermarking systems.