A Multi-Scale Feature Selection Method for Steganalytic Feature GFR

Abstract

The Rich Model of the Gabor filter (referred to as the GFR steganalytic feature) can detect JPEG-adaptive steganography objects. However, feature dimensionality that is too high will lead to too much computation and will correspondingly reduce the detection efficiency. To reduce the dimensionality and the operating time of GFR steganalytic features and to improve the stego image detection accuracy, this paper proposes a multi-scale feature selection method for steganalytic feature GFR. First, we use the SNR criterion to measure the uselessness of each feature component and to provide a basis for the removal of useless steganalytic feature components. Second, we improve the Relief algorithm to measure the importance of feature components in detecting stego images, which provides a basis for the selection of important feature components. Then, we set the threshold value for deleting the useless feature components, and we select the important feature components as the final feature. Finally, we conduct experiments on feature selection for GFR with high-dimensional steganalytic features, and we compared the proposed method with the Fisher-based method, the PCA-based method, the SSFC method, and the Steganalysis-α method. The results show that the method proposed in this paper is effective and fast.

Keywords

• Fast feature selection
• GFR
• steganalytic features
• multi-scale