Convolutional Neural Network-Based Image Watermarking using Discrete Wavelet Transform

TensorFlow 2.5.0 Implementation of "Convolutional Neural Network-Based Image Watermarking using Discrete Wavelet Transform"

Prerequisite

1. Install Python packages

pip install -r requirements.txt

2. Download COCO dataset. Add train, validation, and test images to directories ./train_images, ./validation, and ./test_images respectively. You can change the paths at the configs.py file.

3. Set the models output path at the configs.py file.

Training

To run training:

python trainer.py

By default, it will save a model checkpoint every epoch to MODEL_OUTPUT_PATH. For more arguments and options, see configs.py.

Evaluation

A notebook prepared for evaluation. To run the jupyter notebook use the script bellow:

jupyter notebook

After accessing notebook, open the evaluator.ipynb and run the desire cells.

Reference

If you use this code as part of any published research, please refer the following paper.

@article{tavakoli2023convolutional,
  title={Convolutional neural network-based image watermarking using discrete wavelet transform},
  author={Tavakoli, Alireza and Honjani, Zahra and Sajedi, Hedieh},
  journal={International Journal of Information Technology},
  year={2023},
  publisher={Springer}
}