We are happy that our paper titled “Blind Deep-Learning-Based Image Watermarking Robust Against Geometric Transformations” was accepted at the IEEE International Conference on Consumer Electronics (ICCE) 2024, held in conjuction with the Consumer Electronics Show (CES).
Hannes Mareen will present this research on 6-8 January 2024 at ICCE in Las Vegas, Nevada, USA.
Image watermarking enables protection against copyright infringement. Although existing methods embed watermarks imperceptibly and demonstrate robustness against attacks, they typically lack resilience against geometric transformations.
Therefore, this paper proposes a new watermarking method that is robust against geometric attacks. The proposed method is based on the existing HiDDeN architecture that uses deep learning for watermark encoding and decoding. We add new noise layers to this architecture, namely for a differentiable JPEG estimation, rotation, rescaling, translation, shearing and mirroring.
We demonstrate that our method outperforms the state of the art when it comes to geometric robustness. For example, our model can still correctly decode 77% watermark bits after the watermarked image is cropped to only 25x25 pixels, 10% of its original size (as visualized in the picture below). In general, the bit accuracy remains close to 100% for crops with a ratio of 20% and more.
Example of cropping attack. Even after a 10% crop attack, the watermark can still be 77% correctly decoded.
Additionally, the embedded watermarks are imperceptible - comparable to RivaGAN, a state-of-the-art deep-learning-based watermarking method.
Examples of original (unwatermarked) images, corresponding proposed watermarked images and state-of-the-art RivaGAN watermarked images, as well as their corresponding differences. Although RivaGAN shows less differences, the proposed watermarked images are also imperceptible.
In conclusion, the proposed method can be used to protect images when viewed on consumers’ devices.
Paper: Blind Deep-Learning-Based Image Watermarking Robust Against Geometric Transformations