Prior Classification of Stego Containers as a New Approach for Enhancing Steganalyzers Accuracy

摘要

We introduce a novel "prior classification" approach which can be employed in order to enhance the accuracy of stego detectors as well as to estimate it more subtly. The prior classification is intended for selection a subset of a testing set with such a property that a detection error, calculated over this subset, may be substantially lower than that calculated over the whole set. Our experiments demonstrated that it is possible to select about 30% of the BOSSbase images for which HUGO 0.4 bpp is detected with the error less than 0.003, while the error over the whole set is 0.141. We also demonstrated that it is possible to find about 5% of the BOSSbase images which provide the detection error for HUGO 0.1 bpp less than 0.05, while the error, calculated over the whole set, is about 0.37 which is not quite a reliable accuracy.