Strategy to improve the accuracy of convolutional neural network architectures applied to digital image steganalysis in the spatial domain

In recent years, Deep Learning (DL) techniques applied to steganalysis have surpassed the traditional two-stage approach by unifying feature extraction and classification in a single model, the Convolutional Neural Network (CNN). Several CNN architectures have been proposed to solve this task, improving the detection accuracy of steganographic images, but it is not clear which computational elements are relevant. Here we present a strategy to improve accuracy, convergence, and stability during training. The strategy involves a preprocessing stage with Spatial Rich Models (SRM) filters, Spatial Dropout, Absolute Value layer, and Batch Normalization. Using the strategy improves the performance of three steganalysis and two image classification CNN's, by enhancing the accuracy from 2% up to 10% while reducing the training time to less than 6 hours and improving the networks' stability.

The architecture in the top represents original Ye-Net, the one in the bottom
represents the architecture using the strategy

Citing

If you use our project for your research or if you find this paper and repository helpful, please consider citing the work:

Tabares-Soto R, Arteaga-Arteaga HB, Mora-Rubio A, Bravo-Ortíz MA, Arias-Garzón D, Alzate Grisales JA, Burbano Jacome A, Orozco-Arias S, Isaza G, Ramos Pollan R. 2021. Strategy to improve the accuracy of convolutional neural network architectures applied to digital image steganalysis in the spatial domain. PeerJ Computer Science 7:e451 https://doi.org/10.7717/peerj-cs.451

@article{10.7717/peerj-cs.451,
 title = {Strategy to improve the accuracy of convolutional neural network architectures applied to digital image steganalysis in the spatial domain},
 author = {Tabares-Soto, Reinel and Arteaga-Arteaga, Harold Brayan and Mora-Rubio, Alejandro and Bravo-Ortíz, Mario Alejandro and Arias-Garzón, Daniel and Alzate Grisales, Jesús Alejandro and Burbano Jacome, Alejandro and Orozco-Arias, Simon and Isaza, Gustavo and Ramos Pollan, Raul},
 year = 2021,
 month = apr,
 keywords = {Convolutional neural network, Deep learning, Steganalysis, Strategy},
 volume = 7,
 pages = {e451},
 journal = {PeerJ Computer Science},
 issn = {2376-5992},
 url = {https://doi.org/10.7717/peerj-cs.451},
 doi = {10.7717/peerj-cs.451}
}

This paper was published as a journal paper in PeerJ Computer Science. (Webpage)

Folders

• CNNs and SRM filters This folder contains the notebooks Xu-Net, Ye-Net, Yedroudj-Net, Vgg16Stego, and Vgg19Stego architectures for steganographic images in the spatial domain. Also, it contains the SRMfilters.npy, which are used in the feature extraction stage of CNNs.

Files

• Ye-Net.svg Comparison of Ye-Net architecture with and without the strategy.

Requirements

This repository requires the following libraries and frameworks:

• TensorFlow
• numPy
• OpenCV
• Matplotlib
• Time
• random
• os

This repository was developed in the Python3 (3.8) programming language.

Package installation

if you don't use google colab, We highly recommend to use and install Python packages within an Anaconda enviroment. To create, execute the command below:

conda create --name Strategy python=3.8

So, activate it

conda activate Strategy 

installed the framework

conda install -c anaconda keras-gpu==2.4.3

Now, install the libraries.

pip install opencv-python
conda install -c conda-forge matplotlib
conda install -c jmcmurray os
conda install -c conda-forge time

Execution

After installing all the Requirements, you must clone the repository using.

git clone https://github.com/BioAITeam/
Strategy-to-improve-CNN-applied-to-digital-image-steganalysis-in-the-spatial-domain.git

If you will use colab, upload the cloned folder to drive, then open the folder "CNNs and SRM filters" and run the notebook of your choice.

if you are going to use your computer, install:

conda install jupyter 

Enter the cloned folder, then enter the "CNNs and SRM filters" folder and run the notebook of your choice.

Note

Before running the notebook, please verify that the file paths are correct.

Databases

The data set used to reproduce the results can be downloaded from this link. Images taken from: BOSS competition and BOWS2.