Audio Steganalysis with CNN
@ Author: Wang Yuntao (Charles_wyt)
@ Email: wangyuntao2@iie.ac.cn
Hope we can have a happy communication.
This project is a tensorflow implementation of recent work, you can also design your own network via this platform.
- CNN-based Steganalysis of MP3 Steganography in the Entropy Code Domain [IH & MMSec 2018, Best Paper Award] [paper (ACM)] [paper (pdf)] [dataset]
Necessary Package
tensorflow-gpu==1.4 or later, numpy, pandas, matplotlib, scikit-image, scikit-learn, filetype, virtualenv, librosa (depend on FFmpeg)
You can use command pip install -r requirements.txt to install all packages mentioned above. If you don't want to change your version of tensorflow, you can use virtualenv to create a new python run environment.
How to use
- install python3.x or Anaconda and add the path into the environment variable (recommand python3.5)
- GPU run environment configure if train the network (optional)
- install all dependent packages mentioned above (open setup/requirements.txt and input "pip install -r requirements" into your cmd window)
- run the code as the example as shows
- use tensorboard to visualize the train process such as the accuracy and loss curve of train and validation. The command is "tensorboard --logdir=/path/to/log-directory"
File description
ID | File | Function |
---|---|---|
1 | src | source code |
2 | paper | the PPT and brief introduction of our recent work |
3 | setup | a requirements.txt in this folder, which is used to install all packages in this system |
4 | jupyter | a folder for jupyter debug |
5 | data_processing | tools which are used for QMDCT coefficients extraction and dataset build |