spam

The SPAM (Subtractive Pixel Adjacency Matrix) model is based on the analysis of changes in noise components of a digital image (container) when embedding messages using theory of Markov processes.

The matrix of transition probabilities obtained in the process of modeling inter-pixel dependencies as a higher-order Markov chain is used as a feature vector for the stegoanalyzer. For more references please see corresponding PDF in docs/ folder.

This repository contains the updated version of SPAM extractor of 1st and 2nd order SPAM features from grayscale .png images. This tool is based on related works by T. Pevny, P. Bas and J. Fridrich.

Fixed issues

1. Fix deprecated leaf() (boost C++ lib) to get filename of current directory.
2. Changing the bit_depth field in libpng lib that holds the bit depth of one of the image channels.
3. Change assigning TSpam12 class members with calling libpng's function png_get_image_width(), png_get_image_height(), e.g.

Linux compilation instructions

1. Install boost library:

$ sudo apt-get install libboost-all-dev

2. And just simply build project with make.

Windows compilation instructions

1. Download and compile boost from boost.org. Follow instructions on the boost website (getting started) to compile. In commandline run bootstrap.bat, and then run ./bjam.
2. Download and compile libpng, which also depends on zlib. The libpng solution file also includes zlib project, so you just need to download zlib and place it in the directory structure explained below. The libpng readme also mentions this.
3. Put the compiled libraries in the following directory structure (note that the names of the directories have been changed to remove the versioning):

./THIRDPARTY/Boost/
./THIRDPARTY/zlib/
./THIRDPARTY/libpng/

Open the SPAM.sln in visual studio and compile. Note that it has been configured only for win32 Debug and Release configurations.

Usage

To use this feature extractor you need to allocate the input directory to greyscale PNG images (with inputDir argument), as well as specify upper bound on absolute value of differences on 1st order SPAM features (with T1 flag) and 2nd order SPAM features (with T2 flag):

usage: spam inputDir [-T1] [-T2] [--oFile1st] [--oFile2nd]

For example, to get 1st order SPAM features based on img_cover as input directory to greyscale PNG images call:

$ ./spam img_cover/ --oFile1st features/img_cover_features_1st.txt 

The output would be:

7.0548651550e-02 8.5344252452e-02 8.6745859358e-02 7.9413126534e-02 ... 00001.png
3.1352702929e-02 5.0772029679e-02 7.6292191638e-02 1.0801989712e-01 ... 00002.png
8.4047945956e-02 9.1410392438e-02 9.4881927207e-02 8.8133303397e-02 ... 00003.png
4.0026460292e-02 5.4879773949e-02 7.2883518260e-02 9.2701054888e-02 ... 00004.png
5.9828274841e-02 7.2894042238e-02 8.3003773491e-02 9.5885689419e-02 ... 00005.png

As was in the previous example, to get 1st order SPAM features based on img_stego as input directory to greyscale PNG images call:

$ ./spam img_stego/ --oFile1st features/img_stego_features_1st.txt