Skip to Main Content
Features extraction is the steganographic images (stego- images) classification detection key. How to extract high susceptibility statistic features to noise jamming is a very important thing. A new effective features extraction method was proposed, aiming at the steganographic model with additive noise being the main focus. With the theoretic and experiment analysis, it revealed the difference of principal component egien-values between cover- image and stego-image, and envelope analytic signals with abundance of noise information and high sensitivity to random noise can be extracted associating discrete wavelet transform (DWT) with Hilbert transform (HT). Therefore, multi-scale high frequency sub-bands envelope analytic signals based on DWT/HT were extracted to define sensitive effective feature sets of signals using principal component analysis (PCA). Then built the statistical features extraction model, extracted information entropy and covariance of the images, and constructed sensitive feature vector based on this model. This feature vector was used to detect hidden messages. The simulation result proved its efficiency: sensitivity and separability.