F plus KS: A new feature selection strategy for steganalysis

Performance of universal steganalysis highly depends on the features extracted from the images. Recently there have been some high-dimensional feature sets introduced in order to model a large number of dependencies between neighboring pixels and JPEG coefficients. Although using these high-dimensional models can increase detection rate, due to their dimensionality, they can induce some problems in the classification process. Furthermore, extraction of such excessively large models is time-consuming. Using a feature selection strategy can lead to selection of the most prominent features and as a result, it can decrease feature extraction time. Another advantage of feature selection can be detection of the features that should be preserved in the steganography process in order to avoid detection of steganography. In this paper, a new feature selection algorithm is suggested which utilizes two statistical measures (i.e., KS from Kolmogorov-Smirnov test and F from F-to-remove). For selecting features, the proposed method does not benefit from a classifier; therefore, it should be considered as a filter method. In the proposed method, according to F statistic which is available in F-to-remove method, a reordering is applied on the features. Afterward, the features are mutually compared using KS-test and if the distributions of the two features are equal, one of them is discarded. The comparison of the proposed method with a recently introduced filter-type method for this aim shows performance improvements in terms of the effectiveness of selected features.