Feature based steganalysis, an emerging branch in information forensics, aimsat identifying the presence of a covert communication by employing thestatistical features of the cover and stego image as clues/evidences. Due tothe large volumes of security audit data as well as complex and dynamicproperties of steganogram behaviours, optimizing the performance ofsteganalysers becomes an important open problem. This paper is focussed at finetuning the performance of six promising steganalysers in this field, throughfeature selection. We propose to employ Markov Blanket-Embedded GeneticAlgorithm (MBEGA) for stego sensitive feature selection process. In particular,the embedded Markov blanket based memetic operators add or delete features (orgenes) from a genetic algorithm (GA) solution so as to quickly improve thesolution and fine-tune the search. Empirical results suggest that MBEGA iseffective and efficient in eliminating irrelevant and redundant features basedon both Markov blanket and predictive power in classifier model. Observationsshow that the proposed method is superior in terms of number of selectedfeatures, classification accuracy and computational cost than their existingcounterparts.
展开▼
