Spatial-frequency Feature Vector Fusion Based Steganalysis

This paper presents an algorithm for breaking the JPEG based steganographical algorithms such as F5, one of the most robust information hiding systems. The detection technique is based on spatial-frequency feature vector fusion and SVM classification. First, the proposed method extracts features from spatial domain and DCT domain respectively. Second, the data fusion technique is employed to combine their features. Finally, SVM is used to classify the stego images and non-stego images based on the combined features. Owing to the unique concatenation of two domain features, the proposed algorithm shows high sensitivity to the secret messages of small sizes, allowing a more effective attack.