Research On Blind Detection For Digital Stego Images

Steganography and steganalysis are important issues in the field of information security during the era of Internet.They are drawing more and more attentions from governments, military, security agents and research institutes all over the world. Steganography is an important branch in information hidding, it conceals the true purpose of communication and the fact that the communication occurred through information hidding technology in the public carrier transimission. Steganalysis is developed with steganography, the primary task of steganalysis is to make statistical analysis of the multimedia signals and determine whether the secret message has been embedded in. It has received a great deal of attention both from the academic research and the law enforcement. On the one hand, steganalysis can improve security of steganographic algorithm. On the other hand, it can use to reveal the illegal communication with some steganographic tools. In this thesis, we discussed the blind detection (universal steganalysis) for images.After we analyze popular steganography methods for images and the statistical variance of the cover image brought by the data emdedding process, we propose a new image steganalysis scheme which employs statistical pattern recognition methods to treat the blind detection, and we take the statistical moments in the frequency domain of wavelet coefficient histogram for the cover images and the prediction-error images. Also, we propose to further decomposition of the first scale diagonal D1 subbands and obtain the moments in frenquency domain of the wavelet coefficients histogram of the detail and approximation components for the cover images and the prediction-error images. We prove theoretically and experimentally the effectiveness of the proposed features set. The discrete K-L transform method is used to reduce dimensions, and the BP neural network based on L-M (Levenberg-Marquardt) algorithm is adopted to train the feature set.The experiment results show that the proposed steganalysis scheme is obviously superior to the prior art.Also in this thesis we studied the impact of noise on the detection process, because of the image preprocessing the noise was brought in, and the secret message embedding also can be treated as noise, so in the process of denoising for keeping the sensitivity of the statistical characteristics for the embedded secret message, and the removal of the noise due to image preprocessing is a difficult task, the paper adopted an improved weighted mean of wavelet coefficients for denosing , comparing to other denoising algorithms, the mean square error of the image Significantly reduced, and the PSNR of the image also have different degrees of increase, so this algorithm can be used in denoising before the detection.