An Open Source, MIT licensed implementation of TrustMark watemarking for the Content Authenticity Initiative (CAI) as described in:

TrustMark - Universal Watermarking for Arbitrary Resolution Images

https://arxiv.org/abs/2311.18297

Tu Bui ¹, Shruti Agarwal ² , John Collomosse ^1,2

¹ DECaDE Centre for the Decentralized Digital Economy, University of Surrey, UK.
² Adobe Research, San Jose CA.

This repo contains a Python (v3.8.5 or greater) implementation (python/) of TrustMark for encoding, decoding and removing image watermarks.

Within the python folder run pip install .
The TrustMark package is also available via PyPi - run pip install trustmark to install it indepedent of this repo.

The python/test.py script provides examples of watermarking images (a JPEG and a transparent PNG image are provided as examples). To test the installation the following code snippet in Python shows typical usage:

from trustmark import TrustMark
from PIL import Image

# init
tm=TrustMark(verbose=True, model_type='Q')

# encoding example
cover = Image.open('ufo_240.jpg').convert('RGB')
tm.encode(cover, 'mysecret').save('ufo_240_Q.png')

# decoding example
cover = Image.open('ufo_240_Q.png').convert('RGB')
wm_secret, wm_present, wm_schema = tm.decode(cover)
if wm_present:
  print(f'Extracted secret: {wm_secret}')
else:
  print('No watermark detected')

# removal example
stego = Image.open('ufo_240_Q.png').convert('RGB')
im_recover = tm.remove_watermark(stego)
im_recover.save('recovered.png')

Models are now fetched on first use, due to the size and number of model variants

If you find this work useful we request you please cite the repo and/or TrustMark paper as follows.

@article{trustmark,
  title={Trustmark: Universal Watermarking for Arbitrary Resolution Images},
  author={Bui, Tu and Agarwal, Shruti and Collomosse, John},
  journal = {ArXiv e-prints},
  archivePrefix = "arXiv",
  eprint = {2311.18297},
  year = 2023,
  month = nov
}