Research On Efficient Steganographic Coding And JPEG Cost Function Enhancements

With the rapid development of information technology and the Internet,a large number of sensitive and confidential data,including commercial,military and personal privacy,are transmitted on the network,so the security of data communication is be-coming increasingly important.Digital steganography is a kind of information hiding technology to achieve covert communication or covert storage.By hiding private data in digital multimedia files(e.g.,images,audio and video,text),it can complete the pub-lic transmission or covert storage of secret carriers in a way that does not cause visual/auditory and statistical changes.From the perspective of behavioral security,it thus can avoid the spying and destruction of malicious third parties,and effectively ensures the confidentiality of privacy data.As the most popular image format on the Internet,JPEG image has become an important cover object in steganography.Steganalysis is the coun-termeasure technology to detect steganography.High dimensional steganalytic features and deep learning-based steganalyzer are becoming increasingly powerful,which chal-lenges the steganographic security of JPEG images.To improve the empirical security,JPEG image steganography has developed to the stage of adaptive steganography,aim-ing at minimizing the impact of modifications caused by message embedding.The current adaptive steganographic schemes are generally designed on the min-imal distortion model of "STC coding+cost function".The cost function is used to define the modification cost of each cover element,and the coding method can imple-ment message embedding while minimizing the total cost of modifications.Although STC provides a near-optimal adaptive steganographic codes,from the perspective of the key role of underlying coding technology,only the STC option for adaptive stegano-graphic codes has a potential threat to the development of steganography;and with the growing size of JPEG image files,the implementation of STC embedding on long cov-ers will result in the problem of excessive time consumption.On the other hand,the current JPEG cost functions are additive and symmetric in defining the modification costs of DCT coefficients,which do not properly utilize the neighborhood correlation of natural images and the useful side-information in image processing.To solve the aforementioned problems,this dissertation focuses on three key as-pects in designing steganographic schemes:steganographic codes,fast embedding and cost function.Starting from a JPEG image,the modification costs of DCT coefficients are defined by non-additive cost function based on block boundary continuity or asym-metric cost function based on side-information estimation,the cover length is reduced by cover-shortening algorithm,and the message embedding is completed by stegano-graphic codes based on polar codes.Obviously,the researches in this dissertation can finally generate a series of non-additive/asymmetric fast JPEG steganographic schemes using steganographic polar codes.The main work and innovation of this dissertation are summarized as follows.1.Near-optimal Steganographic Codes based on Polar CodesPolar codes are the only provably channel capacity achieving error-correcting codes with low complexity encoding and decoding methods.Based on the equivalence rela-tionship between the steganographic channel and binary symmetric channel,the parity-check matrix and the decoding initial channel metric are calculated by using the stegano-graphic payload and the steganographic costs.Then an adaptive steganographic coding method is designed based on Successive Cancellation List decoding method,and three strategies are proposed to generalize the codes for arbitrary cover length.Experimental results show that the proposed steganographic codes are suitable for any cover length and any cost function.Compared with STC,it can approach the theoretical bound of steganographic cost with lower embedding time.This near-optimal codes increase the diversity of adaptive steganographic codes and breaks through the dependence of de-signing steganographic schemes on STC.2.Fast JPEG Embedding Algorithm based on Shortening the CoverTo meet the real-time communication requirement of long cover steganography,STC suggests to accelerate the embedding process by decreasing the constraint height h,but this damages the steganographic security to a certain extent.From the perspective of shortening the cover(i.e.reducing the cover length n),properties of shortening the cover,such as shortening the cover will increase steganographic cost and the optimality of constructing the shortened cover,are investigated and proved.According to these properties and the distribution characteristics of DCT coefficients,four shortened cover construction algorithms are designed based on quantization-step partition,cover fold-ing and approximate cost synchronization.Experimental results show that the shortened cover construction algorithm can reduce the embedding time exponentially while main-taining the steganographic security.The cover-shortening based fast JPEG embedding algorithm is a technology connecting steganographic codes and cost functions,so it is suitable for any steganographic codes and any cost function.3.Non-additive JPEG Cost Function based on Block Boundary ContinuityDefining non-additive cost,that is,to consider the interaction between different modifications,is one of the open problems in steganography.Existing JPEG cost func-tions are based on additive distortion model,in which the modifications of DCT coeffi-cients are assumed to be independent and the interactions between different DCT mod-ifications are ignored.To maintain the neighborhood correlation of natural image,the influence on the spatial block boundary of joint modifications of inter-block coefficients with the same DCT frequency is analyzed.Then the principle of block boundary conti-nuity is proposed to define the non-additive costs of horizontal and vertical inter-block DCT coefficients.Under the guidance of this principle,some inter-block coefficients are encouraged to be modified in the same direction,while others are encouraged to be modified in the opposite direction.Experimental results show that the non-additive cost function can enhance the steganographic security of any additive cost function,and it can effectively restrain blocking artifacts caused by steganographic modifications.The principle provides an effective method to calculate non-additive JPEG cost,which lays a foundation for subsequent research.4.Asymmetric JPEG Cost Function based on Side-information EstimationWhen having the spatial image before JPEG compression,side-informed JPEG steganography uses the rounding errors(i.e.side-information)in the JPEG compres-sion process to modulate the modification costs of DCT coefficients,so as to achieve high-level steganographic security.However,in the practical scenario,the steganogra-pher only has one JPEG image without the original spatial image,so it is unable to obtain the side-information.From the perspective of estimating the side-information,a gen-eral framework of side-information estimation based JPEG steganography is proposed.With the deblocking methods in the field of image processing,the side-information can be estimated.Based on observations of the key minorities in cost modulation,five asym-metric cost modulation strategies using the polarity and amplitude of rounding error are carefully designed.Experimental results show that the asymmetric cost function can significantly enhance the steganographic security of any symmetric cost function,and it can even outperform side-informed steganography in some cases.Side-information estimation based steganography with high-level security promotes the development of JPEG image steganography with large capacity,and its framework motivates subse-quent research on asymmetric JPEG cost functions.