The 2nd BOWS Contest: Wavelet-Based Bayesian Denoising Attack on Image Watermarking

In this paper, we present a Bayesian wavelet-based denoising attack on image watermarking. Our approach follows two steps. In the first step, we add a white Gaussian noise to the watermarked image and then apply the Maximum a Posterior Bayesian denoiser in wavelet-domain using Bessel K Form prior. The goal is to obtain a denoising image without watermark. In the second step, we use a basic treatment in the spatial domain to exploit the information contained in the difference image between the marked and the denoising image. The aim is to improve the visual quality of the attacked image and to get a better PSNR. Finally, we present the results obtained in the 2nd BOWS (Break Our Watermarking System) contest to show the effectiveness of our attack.