Research On Distortion-Minimization Steganography And Steganalysis

Steganography and steganalysis are important topics in the field of information security.Steganography can solve information security problems from the root of information,while steganalysis can attack steganography and prevent the spread of illegal information.They play an important role in maintaining national security and social stability.In recent years,the distortion-minimization steganography based on "distortion function + steganographic coding" has become a research hotspot,which can effectively enhance the undetectability of information.The steganalysis technology mainly focuses on traditional machine learning and deep learning.The machine learning method is represented by the spatial rich model and has high steganographic detection performance.This thesis proposes improvements on the basis of the research of previous achievements.The main research work includes:1.Aiming at the problem that the existing distortion-minimization steganographic methods can not completely avoid the smooth region and the clean edge region in the steganography process,an image texture complexity detection model based on multivariate Gaussian carrier model is established to locate the complex texture region of the image.After then,this thesis heuristically define a distortion functions and propose a new spatial adaptive steganography.Firstly,assuming that the steganalyzer can completely estimate texture complexity of the pixel,the likelihood ratio test is used to estimate the texture complexity of each pixel of the cover image,and is associated with the embedded suitability.Then,the embedded distortion value is generated by filtered using a mean filter.Finally,the embedding of secret information is simulated optimal embedding.2.Aiming at the problems of high feature dimension,strong redundancy and long feature extraction time in the steganalysis of spatial rich model,this thesis carries out a series of feature selection operations,and explores the following four different model combination methods: single quantization factor combination method,single filter kernel type combination method,mixed selecting combination method,the Pearson correlation coefficient combination method,and the feature preprocessing means are added.Visual experiments show that our proposed steganography can concentrate the embedded modification into the texture-rich region.Moreover,statistical steganalysis experiments show that our method has better resistance to detection the spatial rich model and its variant version.In addition,two well-known steganography are detected by using the four new feature combinations.Experiments can be obtained: the mixed selecting combination method,the Pearson correlation coefficient combination method can be stable with the performance of steganalysis before dimension is reduced,and have a higher performance dimension ratio.