Image Steganography And Steganalysis Based On Non-additive Embedding

Steganography is a technology of covert communication and JPEG is a common compression format for digital images,what’s more,the current mainstream steganographic algorithm for JPEG images is the adaptive steganographic algorithm.The distortion cost of the existing JPEG image adaptive steganography method is mostly additive distortion,while the non-additive distortions in JPEG steganography have not been sufficiently explored.Steganalysis is a technology resistant to steganography and its research field is mainly focused on general steganalysis.General steganalysis determines whether the image contains secret information by constructing the features of the image before and after steganography,and then using a classifier to train and detect the sample image.The SRM(Spatial Rich Model)feature is a highdimensional feature based on the spatial image residual calculation.It has a good performance when detecting images embedded with additive distortion,but has a low accuracy when detecting images embedded with non-additive distortion.In order to solve the above problems,we study the image steganography and steganalysis based on non-additive embedding in this paper.The research results obtained are as follows:At first,we study the blockiness evaluation method of JPEG images.The blocking artifact is a discontinuous phenomenon that occurs at the block boundary of JPEG image in spatial domain due to JPEG lossy compression.For the stego images generated by JPEG steganography,the embedding changes of DCT(Discrete Cosine Transform)coefficients will affect the blocking artifacts in spatial domain.In order to accurately assess the impact intensity,we find Three typical blockiness evaluation methods.Through experiments,we find that the modification of the DCT coefficients will make the blocking artifacts more serious in spatial domain,and the more secure the steganography algorithm,the less obvious the blocking artifacts.Therefore,if the JPEG steganography reduces the blocking artifacts in the spatial domain when embedding messages,the security of the steganography can be improved.Secondly,we propose a new non-additive distortion design method,which uses the BAR(Blocking Artifacts Reduction)principle to measure the impact of the modification of DCT coefficients on spatial pixels.The main idea of the BAR principle is to guide the selection of the polarity of embedding changes for DCT coefficients in the JPEG domain by reducing blocking artifacts.Because the changes of DCT coefficients will increase the blocking artifacts,the BAR principle can maintain the spatial continuity in both inter-blocks and intra-blocks.Experimental results show that the principle of BAR can be applied to the current adaptive JPEG steganography with additive distortions,and can improve their steganographic security against the advanced JPEG steganalysis,especially at high embedding payloads.Finally,we propose a new method of feature construction based on SRM features.When the non-additive CMD(Clustering Modification Directions)steganography embeds secret information,it will divide the original image into four sub-images,and then dynamically adjust the distortion cost,which can make the modified direction of adjacent pixels consistent and reduce the residuals between pixels.So the non-additive CMD steganography has a good performance when resisting the detection of SRM features.However,the residuals of the four sub-images have not decreased,so we extract the SRM features of the four sub-images to increase the residuals of the original image,and merges them with the SRM features of the original image to obtain a new feature SRM-SS.Experimental results show that the SRM-SS feature constructed in this paper is more accurate than the SRM feature when detecting images embedded in the non-additive CMD algorithm.In summary,we explore the relationship between DCT coefficients modification of JPEG image and blocking artifacts in spatial domain.In addition,we propose a new non-additive distortion design method for JPEG images,which improves the security of existing JPEG steganography with additive distortions.At the same time,we also propose a feature construction method for non-additive distortion embedded images,which improves the detection rate of the non-additive CMD steganography.