We propose SEAL, a watermarking method that embeds semantic information about the generated image directly into the watermark, allowing for a distortion-free watermark that can be verified without a database of key patterns. Instead of relying on stored keys, SEAL infers the key from the image’s semantic embedding using locality-sensitive hashing. Additionally, we address two often-overlooked attack strategies: (i) an attacker extracting the initial noise to create a new image with the same pattern and (ii) an attacker inserting an unrelated, potentially harmful object into a watermarked image while preserving the watermark.
Install all required dependencies by running:
chmod +x setup.sh
./setup.shpython SEAL.pyThe repository contains the code for all experiments discussed in the paper. Explore the provided implementations to reproduce our results and evaluate the robustness of our method against different attacks.
If you find this work useful for your research, please consider citing our paper:
@article{arabi2025seal,
title={SEAL: Semantic Aware Image Watermarking},
author={Arabi, Kasra and Witter, R Teal and Hegde, Chinmay and Cohen, Niv},
journal={arXiv preprint arXiv:2503.12172},
year={2025}
}