LAI, QIRAN and Bors, Adrian orcid.org/0000-0001-7838-0021 (2026) Vector Quantization-Based Watermarking for Autoregressive Generated Images. In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, pp. 9492-9496.
Abstract
Watermarking is essential for identifying AI generated media, while ensuring their copyright protection and proper usage. Existing image generative AI watermarking methods are designed mainly for Generative Adversarial Networks (GANs), or diffusion models, and not for AutoRegressive (AR) models. We propose the Vector Quantization based Watermarking for Autoregressive Generated Images (VQAGI-Mark), a self-attention based image watermarking approach, that selects visually similar token pairs from the codebook and embeds the watermark information through corresponding index replacements, while preserving image quality. Watermarking requires a challenging balancing among its requirements of invisibility, robustness, high bit embedding and security. The proposed watermarking approach embeds multiple bits while enhancing watermark’ robustness. The multi-bit embedded watermark is resilient to a wide range of attacks. Experiments show that the proposed watermark achieves state-of-the-art autoregresive generative watermark fidelity and robustness for watermark identification while maintaining high visual quality. Code is available at https://github.com/TuTAly/VQAGI-Mark
Metadata
| Item Type: | Proceedings Paper |
|---|---|
| Authors/Creators: |
|
| Copyright, Publisher and Additional Information: | This is an author-produced version of the published paper. Uploaded in accordance with the University’s Research Publications and Open Access policy. |
| Dates: |
|
| Institution: | The University of York |
| Academic Units: | The University of York > Faculty of Sciences (York) > Computer Science (York) |
| Date Deposited: | 06 Nov 2026 12:07 |
| Last Modified: | 06 Nov 2026 12:07 |
| Published Version: | https://doi.org/10.1109/ICASSP55912.2026.11464962 |
| Status: | Published |
| Publisher: | IEEE |
| Series Name: | International Conference on Acoustics, Speech and Signal Processing (ICASSP) |
| Identification Number: | 10.1109/ICASSP55912.2026.11464962 |
| Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:247651 |

CORE (COnnecting REpositories)
CORE (COnnecting REpositories)