In order to improve the detection accuracy of hidden message in images, steganalysis features are selected as inputs for steganalysers. However, the existing Fisher criterion ignores the contribution of steganalysis feature components in dispersion to classification, which causes the useful feature components to be deleted, and decreases the detection accuracy of steganalysis features. By analyzing the separability of steganalysis feature components, we introduce the sigmoid function into Fisher's criterion and propose an improved Fisher criterion (I-Fisher criterion), which can make up for the traditional Fisher criterion in separability measurement of steganalysis feature components. To optimize the steganalysis feature and reduce its dimension, we employ the improved Fisher criterion as the heuristic function of the decision rough set α-positive region reduction, and propose the feature selection method based on the improved Fisher. Experimental results show that the proposed method can reduce the dimension and memory of the GFR high-dimensional feature and the CC-PEV lowdimensional feature while maintaining or improving the detection accuracy.
Keywords: I-Fisher criterion; feature selection; separability; steganalysis.