A Robust Reversible Watermarking Scheme Using Attack-Simulation-Based Adaptive Normalization and Embedding

For copyright protection and perfect recovery of the original image in case of no attacks, it is necessary to develop robust reversible watermarking (RRW) methods that counteract both common signal processing (CSP) and geometric deformation (GD) attacks (RRW-CG). However, to the best of our knowledge, none of the existing RRW methods exploit target attacks as prior knowledge to improve their robustness and embedding capacity. To this end, we propose a two-stage RRW-CG scheme with attack-simulation-based adaptive normalization and embedding. Specifically, the polar harmonic transform (PHT) moments are taken as watermark carriers, and their stability with respect to target attacks is evaluated by performing attack simulation tests on large-scale images. This enables the adaptive normalization of PHT moments to improve the watermark robustness. The PHT moments with high stability are then chosen as watermark carriers, and the conventional spread transform dither modulation (STDM) with one quantization level is optimized to form the enhanced version with multiple quantization levels, in which the embedding strength is determined adaptively via attack simulation tests on the candidate watermarked image. This in turn improves the watermark robustness and increases the embedding capacity. After the robust watermark has been embedded, errors caused by robust watermarking are used as the auxiliary information and then inserted into the robustly watermarked image via the recursive code-based reversible watermarking technique, ensuring the reversibility in case of no attacks. Extensive experimental simulation results show that the proposed scheme outperforms the state-of-the-art RRW methods in terms of robustness against CSP such as AWGN, JPEG, JPEG2000, mean filtering, and median filtering as well as GD including rotation and scaling under the same invisibility, reversibility, and embedding capacity. This indicates that, by exploiting target attacks as prior knowledge and designing the attack-simulation-based adaptive normalization and embedding, the proposed novel RRW is feasible and effective.