Unauthorized Detection of CT Watermarks Based on Pattern Analysis Methods

MSc Thesis (July 2006)
Teng Teng

Abstract

The CT watermark algorithm is a relatively new software watermarking algorithm. To evaluate the robustness of the CT watermark algorithm, we tried to simulate the unauthorized detection of watermark attacks on the CT watermark. The main concern of this thesis is to explore the effect of the watermark size ratio of watermarked programs on the accuracy of attackers’ detection function.

We simulate attackers by using the statistical pattern classification method to detect the existence of a CT watermark. The adversaries’ behaviour is confined by a set of rules proposed by the CT watermark algorithm designers, which I extend. We judge the success of attackers by the false positive and false negative rates of attackers’ detection function.

Results showed that when the watermarked programs that the attackers used to train their detection function and the watermarked programs to be detected both have a high watermark size ratio, the attackers’ detection function will be accurate. So restricting the watermark size ratio of CT watermarked programs should increase their robustness.

We suggest two ways to amend this problem. One way is to use the numeric watermark option in the watermark embedding procedure to reduce the watermark size ratio. Another way is to denote those programs which will cause a high watermark size ratio as being unsuitable to be CT watermarked.