Research On Side-Informed Steganography Based On Image Style Transformation

Steganography is one of the most important branches of information hiding,and its application scenarios includes researches on how to utilize the public digital media to realize covert communication of secret information.Steganography techniques not only have great significance in terms of military,intelligence and national security,but also is more closely related units and individuals,groups,privacy protection.With the rapid development of Internet,images,audio,video and other digital media have been widely used.As a result,considering ordinary digital media as communication carrier of steganography technology has a very broad research prospects.The current mainstream research direction of steganography is based on minimiz-ing the distortion model of adaptive steganography,but this kind of steganography pur-suit that images which have been embedded the message and natural images which have not been processed are indistinguishable.But the reality is that on the network social platform to share all kinds of processed image has become more and more popular.Popular image processing can provide sufficient side information for steganogra-phy.It will become the future development trend of steganography that how to utilize all kinds of side information according to the current image processing operations and parameters,in order to hide the personal behavior of secret communication into the mas-sive data of social network without being detected,and then apply the steganography from the laboratory into the real network social media platforms.Now steganography researches based on image processing is not enough,and the research on the side-informed steganography based on image processing is limited to the simple image processing,such as scale zooming,JPEG compression,etc.which will produced quantitative error that can be used as one kind of side information.However,all kinds of image transformation on social platform become more and more popular.According to different type and degree of image processing,to design side-informed models to improve the classical adaptive steganography algorithm is very necessary and meaningful.Based on the review of the concepts and the development status quo of digital image steganography,this thesis then divide the image style transformation into two kinds of circumstances:small changes and severe changes.And then from the point of view of behavior safety to discuss the two scenarios of side-informed steganography,and put forward corresponding side information models of steganography methods for each scenario.In this thesis,the main content and innovation points are as follows:1.Side-informed steganography based on the perturbation of parameter dur-ing image transformation:Proposes a parameter perturbation side information model of steganography al-gorithm for this situation.Utilize gray linear transformation and gamma transform in image processing,using S-UNIWARD and WOW steganography algorithm,discussing in detail for the influence of steganography and steganalysis due to parameter pertur-bations.After parameter perturbation of image transformation,it can make the feature vector of the cover and stego image distributed more randomly in the feature space.Verifying that the theoretical model can apply the steganography hidden in the image pixels' noise produced by parameter perturbation,and introduce the mismatch factors of images base,so it can enhance the detection difficulty for steganalysis.The experiment results have confirmed that the false-alarm rate will be increased,the missed-alarm rate will be reduced,and the total error detection rate is still higher in steganalysis.Within a certain range,the detection error of steganalysis will increase with the increase of wave parameters.2.Side-informed steganography based on the fluctuation of pixels during style transformation:Proposes a pixels fluctuation side information model of steganography algorithm for this situation.Utilize style transformation based on convolution neural network,using S-UNIWARD steganography algorithm,designing the side information based on the changes of pixel values change,defining measurement functions to capture the fluc-tuations for the pixel values changes.Define cover image steganographic distortion according to the sensitivity to style transformation,and adjust the steganographic em-bedding modified probability,to improve the security of original steganalysis algorithm.Experiments show that style transformation can greatly enhance the applicability of the images for steganography.Taking pixel value fluctuation into account as side informa-tion,adjust the pixel base distortion can keep the false-alarm rate,and at the same time,effectively promote the missed-alarm rate.As a result,the security of steganography algorithm has been improved.