A Workbench for the BOWS Contest

The first break our watermarking system (BOWS) contest challenged researchers to remove the watermark from three given images. Participants could submit altered versions of the images to an online detector. For a successful attack, the watermark had to be unreadable to this detector with a quality above 30 dB peak signal-to-noise ratio. We implemented our experiments in R, a language for statistical computing. This paper presents the BOWS package, an extension for R, along with examples for using this experimental environment. The BOWS package provides an offline detector for several platforms. Furthermore, the particular watermarking algorithm used in the contest is analysed. We show how to find single coefficient attacks and derive high-quality images (62.6 dB PSNR) with full knowledge of the key.