Study On Adaptive Steganography And Steganalysis Of Spatial Image Based On Nonnegative Matrix Factorization

At present,one of the development trends of digital image steganography is the adaptive steganography,which aims to adaptively embed information into the most difficult-to-reason regions with high texture complexity in the cover according to the nature of image,thereby improving steganographic security.The digital image steganalysis focuses on how to design a general feature extraction method with high detection rate.In view of the current research status of digital image adaptive steganography and steganalysis,the main work and research results of this dissertation are as follows:We design a new adaptive steganographic methods based on non-negative matrix factorization(NMF),which achieved higher security than the existing steganographic methods.State-of-the-art adaptive steganography mainly focuses on how to design a reasonable cost function and how to utilize that cost function to achieve embedding in the stego image with the minimal distortion based on syndrome-trellis codes(STC).Because previous adaptive steganographic methods use convolution with filters to obtain the residuals,these methods do not make good use of the textured nature of the image itself in the design of the cost function.In this dissertation,we define a new cost function that uses NMF to predict the image pixels and utilizes the mutual dependencies among the pixels to calculate the costs.We present a novel cost function in which the residuals are not calculated via convolution with constant filters.Experimental results show that our method outperforms most state-of-the-art methods,such as MiPOD,S-UNIWARD,WOW and HUGO-BD,in resisting steganalysis based on the spatial rich model(SRM)and is slightly superior to HILL.We design a residual image calculation method for feature extraction of steganalysis based on NMF.The existing steganalysis feature extraction method can be mainly divided into two steps.First,the residual image is calculated by convolution filtering,and then the co-occurrence matrix of residual image is calculated to obtain the final feature.In this dissertation,we analyze NMF with the knowledge of steganalysis,and obtain the NMF specific application method for obtaining residual image in spatial image steganalysis.Considering the number of pixels used by NMF for prediction and its positional relationship with the predicted pixels,a plurality of sets of residual sub-models for acquiring residual images are designed;and then,combined Local Binary Pattern(LBP)and co-occurrence matrix and other statistical information to design the residual combination method,through the experiment to optimize the parameters in feature extraction.Finally,we compare the performance of the designed features with existing steganalysis feature and analyze the validity of the designed features.In addition,we combine the existing artificially designed spatial steganalysis features with the designed features and analyze the validity and complementarity of each type of features,such as SRM,TLBP proposed in [50],and try to find out the lowest dimension and the most effective combination feature from the existing steganalysis features.