Research On Robust Watermarking Resilient To Desynchronization Attacks

Digital media widely spread along with the booming development of computer sci-ence, Internet of Things, cloud computing and the Internet technology. However, unre-stricted reproduction and convenient manipulation of digital media cause considerable economic losses to the media creators and the content providers. Therefore, digital wa-termarking is introduced to prevent the above infringement.In the past ten years, attacks against image watermarking systems have become in-creasingly complicated with the development of watermarking techniques. Up to now, a watermarking scheme that is robust against desynchronization attacks is still a grand chal-lenge. Desynchronization attacks can be classified into three categories:global desyn-chronization attacks and local desynchronization attacks, and non-invertible desynchro-nization attacks. In order to resist this three kinds of desynchronization, we proposed three watermarking resynchronization schemes in this dissertation.In order to resist global desynchronization attacks, we proposed a new watermark-ing resynchronization scheme based on GIRs (geometric-invariant regions). First, we introduced a novel GIRs detection method that is implemented by robust edge contours extraction, robust corners detection, and the radius selection. Then, we designed a new sector-shaped partitioning method for GIR. The sector-shaped partitioning is invariable to geometric transforms, so the sequence of sectors will not be out-of-order under ge-ometric transforms. The GIRs and the divided sector discs are invariant to geometric transforms, so the watermarking method inherently has high robustness against global desynchronization attacks as shown in experiments.We presented a blind image watermarking resynchronization scheme against local transform attacks. Firstly, we propose a new feature transform named local daisy feature transform (LDFT), which is not only globally but also locally invariable. Then the binary space partitioning (BSP) tree is used to partition the geometrically invariant LDFT space. In the BSP tree, the location of each pixel is fixed under global transform, local transform, and cropping. Lastly, the watermarking sequence is embedded bit by bit into each leaf-node of the BSP tree by using the logarithmic quantization index modulation (LQIM) watermarking embedding method.In order to resist non-invertible desynchronization attacks, we proposed a novel two- stage saliency detection model by fusing bottom-up and up-down features extracted from a single image in this dissertation. The evaluation of the proposed model has been carried out on two largest publicly available data sets. As indicated in the experimental results, the proposed model consistently outperforms13existing saliency detection methods with higher precision and better recall rates. The watermarking resynchronization scheme is proposed based on the salient region of the image. Because the salient region will not be cropped when the image is attacked by non-invertible desynchronization attacks (such as:image retargeting), the proposed the watermarking scheme is robust against the non-invertible desynchronization attack.