Watermark re-synchronization using log-polar mapping of image autocorrelation

Many watermarking algorithms embed the watermark into the image as contiguous non-overlapping tiles. This tiling structure forms an implicit synchronization template that can be revealed through autocorrelation. This template is composed of a set of weak peaks, replicating the relative position of the watermark tiles. Hence, synchronization can be resolved by comparing the actual locations of these peaks to the theoretical ones to determine the scaling factor and the orientation angle of the tiles. Unfortunately, these peaks are very weak and measuring their locations directly is not easy. In this paper, a log-polar mapping of the synchronization template is computed to convert the scaling factor and the rotation angle of the template into vertical and horizontal shifts. These shifts are then detected using a Phase-Only-Matched filter (POM), which concentrates the weak energy from all peaks into a single peak that is much easier to detect. The scaling factor and orientation angle are determined from the location of this peak. Simulation results of this method have shown that this method is very effective and produces accurate results.