Feature Selection Based on Clonal Selection Algorithm for Image Steganalysis

With the rapid development of steganography, researches on the adaptive steganography have made steganalysis more and more difficult. In order to improve the detection accuracy, there are plenty of works introducing complex high-dimensional features to steganalysis, among which the rich model steganalysis features are particularly classical. However, there are many redundant features in high-dimensional features, which greatly increase the time and computational cost. How to eliminate redundant features while maintaining detection accuracy is still a research hotspot. In this paper, we propose a novel feature selection method based on Clonal Selection Algorithm (FS-CSA) for image steganalysis, which separates features into multiple populations according to the idea of niche technique, and proposes a feature similarity function to calculate the affinity of features in each population, and finally uses FS-CSA to select features based on the affinity in each population in turn. The experimental results show that FS-CSA can significantly reduce the feature dimensionality while maintaining the detection accuracy, where the Spatial Rich Model (SRM) features can be reduced to about 8000 dimensions, and the Gabor Filter Residual (GFR) features can be reduced to about 7000 dimensions. The crossover experiments show that the steganalysis features selected by FSCSA are also generalizable to any steganography algorithm.