Improving the robustness of nonadditive watermarks through optimum detection theory

In this paper, a new watermarking system for copyright protection of digital images is presented. The method operates in the frequency domain, by embedding a pseudo-random sequence of real numbers in a selected set of DFT coefficients of the image. Moreover, the masking characteristics of the Human Visual System are exploited for watermark hiding and a synchronization pattern is introduced into the watermarked image to cope with geometrical attacks. By relying on statistical decision theory, a new decoding algorithm which is optimum for non-additive full-frame DFT watermarks is then derived. Experimental results highlight both the superiority of the novel detector scheme with respect to conventional correlation-based decoding, both the robustness of the overall system against a large set of attacks aiming at removing the watermark.