Improved watermark detection reliability using filtering before correlation

A digital watermark is a perceptually unobtrusive signal embedded in some multimedia asset carrying additional information: e.g. copyright information of a video clip. For nearly all watermarking schemes published so far, detection is based on some form of correlation. We show that the detection reliability can be significantly improved by applying filtering prior to correlation. This improvement is analysed using a theoretical model based on statistical communication and detection theory. Finally, the improvements predicted by the theory are verified in a number of experiments.