Research On Steganography Security Models And Universal Steganalysis For Images

Steganography and steganalysis are of great significance to information security in that they become very important techniques on information war in the internet age. Steganography security models and universal steganalysis for images with passive adversaries are focused on in this thesis. The main contributions are summarized as follows.1. Two steganography security definitions fitted for images are proposed from different points of view. One is in terms of the relative entropy based on the reversible de-correlation transform. The other is in terms of the conditional relative entropy using the markov random field model for images. Different image models are discussed for the definitions respectively. Both definitions can be extended to other cover types such as audio, video data.2. A new universal steganalysis algorithm for the additive noise modelable steganography in the spacial domain is put forward. Higher order statistics moments of the differential histogram distributions and the ratios of special areas under the differential histogram curves are calculated to characterize the difference between the differential distribution of the cover image and that of the stego image. Experimental results show that our technique achieves better detection performance than Farid's.3. A new universal steganalysis method is presented for the steganography techniques in the JPEG compression domain. The features for the blind classifier are calculated as the mean, variance, skewness, kurtosis of the histogram distribution of the DCT coefficients, and the histogram distribution probabilities of special coefficient values as well. The variance analysis is utilized to identify useful features in steganalysis. Experimental results prove the validity of the method. Furthermore, an improved version is proposed by introducing the features which characterize the inter-block distributions so that the algorithm can detect stego images efficiently for the steganography preserving the histogram property.4. A new steganorgraphy preserving the histogram property is proposed based on adjusting the modification direction of the coefficients dynamically. Several experimentresults demonstrate that the algorithm can attain a high hiding capacity of 13% (when the quality factor is 75). Furthermore the histogram property is fine changed, and various histogram-based attacks can be effectively resisted.5. A universal steganalysis scheme is presented that principle component analysis is introduced to preprocess the image statistics features and steganalysis classifier is constructed using RBF network. The comparison of the simulation results shows that our scheme is quite more efficient than Farid's scheme because the stego image that the proportion of the embedding message to the maximal embedding capability is more than 60% (Jsteg), 80% (EzStego), 50% (S-Tools) can be detected efficiently by our scheme.6. A new universal steganalysis algorithm is proposed for color images that the features are extracted in view of the luminance and the chrominance of the image. A new approach is presented to extract the feature for the chrominance component that the vector direction correlation between the image and its directional filtered image is calculated. It reduces the number of the features enormously, and overcomes the fault of Farid's scheme that neglects the inherent correlation between the RGB channels. Experimental results demonstrate that our scheme is quite more efficient than Farid's.