Steganalysis by Subtractive Pixel Adjacency Matrix
read more
Citations
Rich Models for Steganalysis of Digital Images
Ensemble Classifiers for Steganalysis of Digital Media
Break our steganographic system: the ins and outs of organizing BOSS
Universal distortion function for steganography in an arbitrary domain
Using high-dimensional image models to perform highly undetectable steganography
References
The Nature of Statistical Learning Theory
A Practical Guide to Support Vector Classication
Feature extraction : foundations and applications
F5-A Steganographic Algorithm
Low-complexity image denoising based on statistical modeling of wavelet coefficients
Related Papers (5)
Frequently Asked Questions (13)
Q2. What have the authors stated for future works in "Steganalysis by subtractive pixel adjacency matrix" ?
In their future work, the authors would like to use the SPAM features to detect other steganographic algorithms for spatial domain, namely LSB embedding, and to investigate the limits of steganography in the spatial domain to determine the maximal secure payload for current spatial-domain embedding methods. Another direction worth pursuing is to use the third-order Markov chain in combination with feature selection to further improve the accuracy of steganalysis.
Q3. What is the heuristic behind embedding by noise adding?
The heuristic behind embedding by noise adding is based on the fact that during image acquisition many noise sources are superimposed on the acquired image, such as the shot noise, readout noise, amplifier noise, etc.
Q4. Why do the authors believe the simple filter is superior to more complex filters?
The authors believe that the superior accuracy of the simple filter [−1,+1] is because it does not distort the stego noise as more complex filters do.
Q5. How can the authors model dependences between pixels in natural images?
In principle, higher-order dependences between pixels in natural images can be modeled by histograms of pairs, triples, or larger groups of neighboring pixels.
Q6. How many bins in an 8-bit grayscale image?
The curse of dimensionality may be encountered even for the histogram of pixel pairs in an 8-bit grayscale image (2562 = 65536 bins).
Q7. What is the effect of the curse of dimensionality?
In their paper, the authors show that there is a great performance benefit in using higher-order models without running into the curse of dimensionality.
Q8. What is the way to improve the accuracy of steganography?
Another direction worth pursuing is to use the third-order Markov chain in combination with feature selection to further improve the accuracy of steganalysis.
Q9. How many features were shared between all four databases?
At the same time, one must be aware that the feature selection is database-dependent as only 114 out of 200 best features were shared between all four databases.
Q10. What is the difference between neighboring pixels?
The local dependences between differences of neighboring pixels are modeled as a Markov chain, whose sample probability transition matrix is taken as a feature vector for steganalysis.
Q11. Why is the feature set susceptible to the curse of dimensionality?
Because the dimensionality of the second-order SPAM feature set is 686, the feature set may be susceptible to all the above problems, especially for experiments on the NRCS database.
Q12. What is the classification error of the steganalyzers?
In all cases, the steganalyzers that used second-order SPAM features perform the best, the WAM steganalyzers are second with about three times higher error, and ALE steganalyzers are the worst.
Q13. How many training samples should be at least ten times the dimension of the training set?
In theory, the number oftraining samples depends exponentially on the dimension of the training set, but the practical rule of thumb states that the number of training samples should be at least ten times the dimension of the training set.