@techreport { VG:VPI2000,
author = { Sviatoslav Voloshynovskiy and Shelby Pereira and Victor Iquise and Thierry Pun },
title = { Attack Modelling: Towards a Second Generation Watermarking Benchmark },
institution = { Computer Vision Group, Computing Centre, University of Geneva },
year = { 2000 },
vgclass = { report },
vgproject = { watermarking },
number = { 00.05 },
address = { rue G\'en\'eral Dufour, 24, CH-1211 Gen\`eve, Switzerland },
month = { may },
abstract = { Digital image watermarking techniques for copyright protection have become increasingly robust. The best algorithms perform well against the now standard benchmark tests included in the Stirmark package. However the stirmark tests are limited since in general they do not properly model the watermarking process and consequently are limited in their potential to removing the best watermarks. Here we propose a stochastic formulation of the watermark removal problem, considering the embedded watermark as additive noise with some probability distribution. The attack scheme consists of two main stages: a) watermark estimation and partial removal by a filtering based on a Maximum a Posteriori (MAP) approach; b) watermark alteration and hiding through addition of noise to the filtered image, taking into account the statistics of the embedded watermark and exploiting HVS characteristics. In a second stage we propose a ``second generation benchmark''. We follow the model of the Stirmark tests and propose the 6 following categories of tests: denoising attacks, wavelet compression, denoising/compression with following perceptual remodulation, template/ACF (synchronization) removal, denoising and random banding, the watermark copy attack. Our results indicate that even though some algorithms perform well against the stirmark benchmark, all algorithms perform poorly against our benchmark. This indicates that much work remains to be done before claims about ``robust'' watermarks can be made. We also propose a new method of evaluating image quality based on the Watson metric which overcomes the limitations of PSNR. }
}