Abstract
Watermark robustness against both geometric and signal processing attacks is still challenging in the watermarking community. In this work, we propose an approach to increase the watermark detection rate. The novel feature of the method resides in pre-processing the image before the embedding process, and involves no interaction with other improvements. It is based on the following idea: given the same watermarking scheme, some images are naturally more appropriate for the embedding process as they obtain better detection rates than others. In the Fourier scheme for example, it was observed that lower amplitude variations at Fourier coefficients correspond to a better detection rate. We theoretically justify these findings and demonstrate that by applying a dedicated low-pass filtering on the magnitude of the frequency bands the watermark detection rate is significantly improved.