[TMM'25] FontGuard: A Robust Font Watermarking Approach Leveraging Deep Font Knowledge

Description

We introduce FontGuard, a novel font watermarking model that harnesses the capabilities of font models and language-guided contrastive learning. Unlike previous methods that focus solely on the pixel-level alteration, FontGuard modifies fonts by altering hidden style features, resulting in better font quality upon watermark embedding. We also leverage the font manifold to increase the embedding capacity of our proposed method by generating substantial font variants closely resembling the original font. Furthermore, in the decoder, we employ an image-text contrastive learning to reconstruct the embedded bits, which can achieve desirable robustness against various real-world transmission distortions.

Training

The training code will be available soon.

Demo

We provide the SVG files for the 1-bit watermarked SimSun font which includes two variants, along with a test set collected across 7 distribution scenarios. Each scenario's test set contains 1000 segmented character images. All the data can be found here (Password: rocu).

The data is organized as follows:

|-WeChat  # test set for OSNs scenario
|---FontGuard_SimSun_16
|-Weibo  # test set for OSNs scenario
|---FontGuard_SimSun_16
|-Whatsapp  # test set for OSNs scenario
|---FontGuard_SimSun_16
|-Facebook  # test set for OSNs scenario
|---FontGuard_SimSun_16
|-print_camera  # test set for cross-media scenario
|---FontGuard_SimSun_16
|-screen_camera  # test set for cross-media scenario
|---FontGuard_SimSun_16
|-screenshots  # test set for cross-media scenario
|---FontGuard_SimSun_16
|-svg  # 1-bit watermarked SimSun SVG
|---msg_0
|---msg_1
|-bit_seq.txt  # bitstream ground truth for the test set
|-dec.pth  # decoder checkpoint
|-GB2312_CN6763.txt  # character set

To extract the bitstream from test set font images, first update the paths in cfg.py. Then, execute the following command:

python test.py

Citation

If you find our project useful in your research, please cite it in your publications.

@article{
  wong2025fontguard,
  title={FontGuard: A Robust Font Watermarking Approach Leveraging Deep Font Knowledge},
  author={Wong, Kahim and Zhou, Jicheng and Li, Kemou and Si, Yain-Whar and Wu, Xiaowei and Zhou, Jiantao},
  journal={IEEE Transactions on Multimedia},
  year={2025}
}