A mishmash of methods for mitigating the model mismatch mess

The model mismatch problem occurs in steganalysis when a binary classifier is trained on objects from one cover source and tested on another: an example of domain adaptation. It is highly realistic because a steganalyst would rarely have access to much or any training data from their opponent, and its consequences can be devastating to classifier accuracy. This paper presents an in-depth study of one particular instance of model mismatch, in a set of images from Flickr using one fixed steganography and steganalysis method, attempting to separate different effects of mismatch in feature space and find methods of mitigation where possible. We also propose new benchmarks for accuracy, which are more appropriate than mean error rates when there are multiple actors and multiple images, and consider the case of 3-valued detectors which also output `don't know'. This pilot study demonstrates that some simple feature-centering and ensemble methods can reduce the mismatch penalty considerably, but not completely remove it.

[1]  Jessica J. Fridrich,et al.  Ensemble Classifiers for Steganalysis of Digital Media , 2012, IEEE Transactions on Information Forensics and Security.

[2]  Mauro Barni,et al.  Forensics aided steganalysis of heterogeneous images , 2010, 2010 IEEE International Conference on Acoustics, Speech and Signal Processing.

[3]  Xiaoyuan Yang,et al.  A New Multi-class SVM Algorithm Based on One-Class SVM , 2007, International Conference on Computational Science.

[4]  Rainer Böhme,et al.  Moving steganography and steganalysis from the laboratory into the real world , 2013, IH&MMSec '13.

[5]  Andrew D. Ker,et al.  Steganalysis with mismatched covers: do simple classifiers help? , 2012, MM&Sec '12.

[6]  Koby Crammer,et al.  A theory of learning from different domains , 2010, Machine Learning.

[7]  Jessica J. Fridrich,et al.  Steganalysis of JPEG images using rich models , 2012, Other Conferences.

[8]  Andrew D. Ker,et al.  Steganalysis using logistic regression , 2011, Electronic Imaging.

[9]  Tomás Pevný,et al.  "Break Our Steganographic System": The Ins and Outs of Organizing BOSS , 2011, Information Hiding.

[10]  Jessica Fridrich,et al.  Kernel methods in steganalysis , 2008 .

[11]  Andrew D. Ker.  Batch Steganography and Pooled Steganalysis , 2006, Information Hiding.

[12]  Tomás Pevný,et al.  The challenges of rich features in universal steganalysis , 2013, Electronic Imaging.

[13]  Bernhard Schölkopf,et al.  Domain Adaptation under Target and Conditional Shift , 2013, ICML.

[14]  Tomás Pevný,et al.  Steganalysis by subtractive pixel adjacency matrix , 2010, IEEE Trans. Inf. Forensics Secur..

[15]  Bernhard Schölkopf,et al.  Estimating the Support of a High-Dimensional Distribution , 2001, Neural Computation.

[16]  D. Munson.  A note on Lena , 1996 .

[17]  Jessica J. Fridrich,et al.  Random projections of residuals as an alternative to co-occurrences in steganalysis , 2013, Electronic Imaging.

[18]  Tomás Pevný,et al.  From Blind to Quantitative Steganalysis , 2012, IEEE Trans. Inf. Forensics Secur..

[19]  Andrew D. Ker,et al.  Exploring multitask learning for steganalysis , 2013, Electronic Imaging.