Research On Key Issues Of Robust Image Steganography For Lossy Network Channels

Digital steganography embeds information into redundant parts of various multimedia carriers,such as images,audio,video,and text,and transmits them over open channels for secure covert communication.It can hide not only the content of the communication but also the existence of secret communication actions,thus becoming one of the important communication means in the special defense and civil fields.In recent years,social media,such as Facebook and We Chat,have become a modern way for people to communicate information.At the same time,images have become a popular medium for people to share information.Image steganography technology towards social network channels has brought more opportunities for digital steganography.However,images transmitted over social network channels usually suffer from compression,scaling,denoising,and other attacks.Traditional steganography techniques focus on anti-statistical detection performance but do not consider the image attacks such as scaling and compression.It is difficult to extract complete information from an attacked image.Therefore,the study of image robust steganography for lossy network channels,which takes into account the anti-statistical detection performance as well as the robustness of anti-scaling and anti-compression,is valuable both in theory and in practice.This dissertation focuses on several key issues of this technique.The main work and innovations of this dissertation are summarized as follows:1.Existing steganography measurements do not accurately evaluate the performance of image robust steganography methods.To address this issue,a five-in-one measurement PRUDA(Payload,Robustness,ease of Use,anti Detection,Applicability)for image robust steganography is proposed.We Chat transmission tests of 1000 scene images taken by 7 top-ranked mobile phones show that the definition of Payload and Applicability is reasonable.Statistical tests on the composition of images shared by 1,917 intelligent mobile users show that the definition of anti Detection is reasonable.Compared to existing measurements that only consider robustness and anti-statistical detection,PRUDA compensates for the lack of measurements such as actual embedding information,available carriers,and adaptation between methods and channels.It is helpful to accurately measure the performance of image robust steganography methods for lossy network channels and guide the design of steganography methods adapted to different channels.2.Existing image robust steganography methods weaken the security of adaptive steganography.To address this issue,an image robust steganography method based on edge enhancement filtering is proposed.Based on the analysis of the security of the asymmetric distortion steganography strategy,a high-quality side-information image with enhanced edges and contours is constructed by combining the filtering operation and the superpixels that delineate the edges and contours of objects.Combining the asymmetric distortion adjustment strategy and adaptive steganography framework,the steganography distortion function that can accurately predict the embedding modification direction is designed,and the distortion function is applied to adaptive steganography and robust steganography.Experimental data based on four databases including Bossbase-1.01,UCID,BOWS2,and IStego100K show that compared to the existing typical adaptive steganography methods J-UNIWARD and UERD,the proposed method improves the anti-statistical detection performance by 7.6%and 6.0%,respectively.Compared to the typical robust steganography method JCRISBE,the proposed method improves the anti-statistical detection performance by an average of 8%.3.Existing robust steganography methods against compression cannot extract information correctly when those methods suffer compression attack with different parameters.To address this issue,a robust steganography method based on the position invariance of DCT coefficients is proposed.Based on the analysis of DCT coefficients in JPEG images before and after compression,the non-zero invariance of DCT coefficients and the position invariance of the last non-zero DCT coefficient in a DCT block are mined.The robust carrier is constructed using the position invariance of DCT coefficients,and the rules for the modification position calculation and the distortion function design are devised.Combined with a minimization embedding framework and error-correcting coding,an image steganography that resist compression attacks and statistical detection is proposed.Experimental results show that compared to existing typical robust steganography methods MREAS-P_Jand JCRISBE,when a cover image quality factor is65 and the difference between a channel attack quality factor and the cover image quality factor is[0,5],the information extraction error rate of the proposed method is reduced by 69%on average.It is also verified that the proposed method can against attacks with different compression parameters.4.Existing anti-scaling robust steganography methods cannot effectively resist common used interpolation attacks.Therefore,a robust steganography method based on inverse interpolation is proposed.Based on the analysis of the scaling principle of bilinear interpolation and bicubic interpolation that are commonly used in image scaling,the mapping rules between pixel regions in an image and pixels in the corresponding scaled image are mined,and the inverse interpolation equations are constructed by combining variable constraints of non-overdetermined equations.The adaptive steganography framework of"distortion function design+minimization embedding"is combined with the adjustment obtained by solving the equations,and the image robust steganography can resist scaling attack and statistical detection is proposed.Theoretical analysis shows that the method can ensure complete extraction of information towards targeted scaling attacks.Experimental results show that compared with the existing typical adaptive steganography methods S-UNIWARD and Mi Pod,and the typical anti-scaling robust steganography methods QIMIS and ZMIS,the average information extraction error rate of the proposed method is 0 when it is subjected to bilinear interpolation scaling attacks with scaling factors of 0.5 and 0.25.5.The robustness of the existing anti-scaling image robust steganography methods needs to be improved when they are subjected to bicubic interpolation scaling attacks.Therefore,a robust steganography method based on an optimal adjustment strategy is proposed.Based on the analysis of the adjustment that is closely related to the information extraction error rate,a mass of candidate embedding positions are mined and the inverse interpolation equations are constructed.Utilizing the error between the approximate solution and the exact solution,the optimal adjustment selection strategies are proposed for different objectives.The optimal adjustment is selected by the proposed max-error priority strategy,which is combined with the robust steganography framework to combat the bicubic interpolation scaling attack.Experimental results show that compared with the typical existing adaptive steganographic methods S-UNIWARD and Mi Pod,and the typical anti-scaling robust steganographic methods QIMIS and ZMIS,the proposed method maintains complete information extraction when toward bilinear interpolation scaling attack,and its information extraction error rate reduces about 70%when toward bicubic interpolation scaling attack with a scaling factor of 0.25.Finally,this dissertation is summarized,and some problems that need further study are pointed out.