Invisible watermark contains two algorithms: significant Bit(LSB) and Discrete Cosine Transform(DCT) refer guofei9987 , importantly , the DCT algorithm needs the carrier image size should 16 times greater than watermark, or it doesn't work.
These scripts need python 3.6+ and the following libraries:
numpy==1.19.5
opencv-python==4.5.1.48
PyWavelets==1.1.1
You can easy to install these libraries by :
pip install -r requirements.txt
First of all, you need satisfy the environment requirements, then run
python main.py --ori_filename <carrier image path> --wm_filename <watermark path>
you can get a image include watermark, the default path under***./img/***.
Above-mentioned is the most convenient way encode, but formally, you need to run this
python main.py --ori_filename <carrier image path> --wm_filename <watermark path> --coder encoder
For example
python main.py --ori_filename ./img/carrier.png --wm_filename ./img/watermark.png --coder encoder
In addition, you can use -wm_method
to choose DCT algorithm (default) or LSB algorithm to encode the watermark, you can just run like this
python main.py --ori_filename ./img/carrier.png --wm_filename ./img/watermark.png --coder encoder --wm_method LSB
Furthermore,it also support the usual attack method to destroy image,you need to run this
python main.py --ori_filename ./img/carrier.png --wm_filename ./img/watermark.png --coder encoder --att_method saltnoise
The all attack methods you can look up in ./src/attack.py, and the watermarked image will be saved under the path *./img/.
Above all, you need to confirm the watermarked image is encoded by which algorithm, then you can run like this
python main.py --wm_filename ./img/watermark.png --ori_wm_filename ./img/ori_wm.png --coder decoder
As you see, the all algorithms needs the watermark image to decode, because the watermark size we don't know.