Optimizing Feature for JPEG Steganalysis via Gabor Filter and Co-occurrences Matrices

摘要

For modern steganography algorithms, there are many distortion functions designed for JPEG images which are difficult to be detected for the steganalyst. Until now, the most successful detection of this kind steganography named GFR (Gabor Filter Residual) is currently achieved with detectors for training on cover and stego sets. These features extract the image texture information from different scales and orientations, and the image statistical characteristics can be captured more effectively. In this paper, we describe a novel feature set for steganalysis of JPEG images. The features are composed of two parts. All of them are obtained based on GFR in the spatial domain. Its first part is to extract the histograms features, and the other part is co-occurrence matrices features. Due to its high dimensionality, we make the best of the label to reduce these features. Compared with state-of-the-arts methods, the most advantage of this proposed steganalysis features is its lower detection error while meeting the advanced steganographic algorithms.