Myung Joon Kim

Watermark design for the Wiener attack and whitening filtered detection

A well-known power spectrum condition (PSC) was derived for resisting the Wiener attack. The Wiener attack estimates the embedded watermark from a watermarked signal and subtracts it to disable the watermark detection. According to the PSC, the power spectral density of a watermark should be proportional to that of the original data since the watermark becomes difficult to estimate. However, PSC considers the correlation-based detection only and does not pay attention to whitening filtering before detection. In watermark detection problem, the host signal is considered as a noise and usually it has a colored power spectrum. Thus, by applying the whitening filter the watermark detection performance can be considerably improved. When using the whitening filter, if the power spectrum of watermark is like that of the original data, the gain of the whitening filter becomes one and there is little enhancement in detection performance. On the contrary, the watermark with different power spectral density from the original data can be removed by the Wiener attack, and the detection of the watermark becomes difficult. So this paper aims to design a watermark to improve the detection performance by satisfying apparently opposite conditions. The watermark optimized with calculus of variation method achieved the above objective in the experiment.