Feature extraction optimization of JPEG steganalysis based on residual images

Abstract

JPEG steganalysis is a technique for detecting the existence of secret information hidden in JPEG images. At present, JPEG steganalysis based on residual images has been widely studied due to its low complexity and good performance. Due to the use of different quality factors in the JPEG image, the features need to be scaled appropriately according to this quality factor to obtain better detection performance. However, the current method of selecting the appropriate scale is usually based on empirical formulas or experiments. To solve this problem, we propose a novel adaptive scale adjustment algorithm for feature extraction filters. This algorithm can adaptively adjust the scale of the feature extraction filter according to the quality factor of JPEG images for better performance. In addition, we also propose a new weighted histogram method. Before calculating the histogram features of the residual image, we do not perform the rounding operation as usual, and different weights are assigned according to the distance between each input and histogram bins. The experimental results on BossBase set show that the proposed method can improve the performance of steganalysis.