Research On Universal Steganalysis Based On Multi-Carrier Image

With the rapid development of the Internet, information security technology has become crucial. The information hiding technology has experienced significant research and development over the last decade, and attracted much attention. There is not only a contradiction but also a complementary relationship between the techniques of steganalysis and steganography.This paper proposes a universal steganalysis method which is independent of image formats and a universal steganalysis method based on RICH model with a small embedding rate. This paper studies the analysis technique using multiple image formats and mainly aims at universal steganalysis method for multi-carriers. There are mainly two parts of research work in this paper.A universal steganalysis method is proposed for detecting general formats of images. This method is based on statistical feature extraction methods of multi-domain features and formed through the study of steganography algorithms in JPEG, BMP and GIF formats. The corresponding features are firstly extracted from contour wave field. Then correlation of the symbiotic discrete cosine transform coefficients is calculated using the joint probability density and distribution of coefficient pairs in the images is calculated using the correlation matrix; they and extended to the airspace and used to extract feature values. The feature values are then extracted from calibrated image using the same method. At last, the resulting difference between the final two features is defined as a feature vector for the classifier training. Experimental results show when compared with "pair analysis method" proposed by Jessica Fridrich and "wavelet packet transform method" proposed by Luo Xiangyang, the proposed method in this paper achieves a better detection accuracy rate of over75%. It works better in steganalysis when performed for multiple image formats.To solve the problem of low detection rate when the embedded rate is small, a novel universal steganalysis method is proposed based on Rich model with a small embedding rate. This method is based on the noise component model and texture component model. First, image features are extracted from the analysis process of wave contour analysis, image de-noising and neighborhood prediction. Then these features are calibrated to reflect changes better after embedding secret information; Second, ensemble classifier is selected to determine whether the image contains hidden information. A predictive image is used to remove the characteristic of the image itself in order to strengthen the classification results. The experimental simulation results show that the detection result has high reliability when the embedded amount is larger than1KB.