<title>Unified approach of asymmetric watermarking schemes</title>

Abstract

Asymmetric schemes belong to second generation of watermarking. Whereas their need and advantage are well understood, many doubts have been raised about their robustness and security. Four di erent asymmetric schemes have been proposed up to now. Whereas they were seemingly relying on completely di erent concepts, they share the same performances. Exploring in detail these concepts, the authors propose a common formulation of the four di erent detector processes. This allows to stress common features about security of asymmetric schemes.

An asymmetric watermarking method

“…Although invented independently, the authors showed that these methods shared the same detection algorithm (i.e. the correlation is replaced by a quadratic form), and that they have similar performances [30]. The only exception is the work from and J. Picard and A. Robert based on neural networks [31].…”

“…This is not the case in actual asymmetric techniques. We produced in [30] an attack available if the detection key is disclosed. This attack threatens all methods unified in this later paper.…”

Optical asymmetric watermarking using modified wavelet fusion and diffractive imaging

“…And then optical asymmetric cryptosystems were proposed for security enhancement [13][14][15][16][17][18][19]. Similarly, various digital asymmetric watermarking schemes have also been generated at the cost of high computational time [20][21][22][23][24][25]. In this regard, it is necessary to develop an asymmetric watermarking technique which should be; (1) non-iterative, (2) able to watermark both: binary as well as gray-scale images, and (3) enhanced level of security and should follow the optical approach employing optical transforms.…”

Secure watermark detection with nonparametric decision boundaries

“…In [3], the authors described the asymmetric watermarking technique that is used a test statistic in a quadratic form. Although the decision boundary is more complicated in this case, it is still parametric and can be estimated, even in the least square sense using a fmite number of samples.…”