Threshold Selection for Correlation-based Watermark Detection

In this paper, the problem of how a decision threshold for a correlation-based watermark detection algorithm can be properly chosen is addressed; it is demonstrated that, if a water-marked image is attacked through signal processing techniques, the statistic of the correlation values changes, resulting in an higher false negative error probability than was expected. To cope with this eeect, a new threshold based on a xed constraint on the maximum probability of false positive errors is proposed. Results connrm the robustness of the decoder against various kinds of attacks.