Research On Steganography And Steganalysis In Digital Videos

Information is a crucial strategic resource,and information security is related to the overall situation of national stability and development.As an important research direction in the field of information security,steganography and steganalysis have received increasing attention.Steganography can ensure the reliable transmission of information,but its illegal application will bring serious security risks.Steganalysis is a countermeasure of steganography.It can effectively identify steganographic behaviors,and has practical significance for strengthening information supervision,obtaining valuable information,and maintaining national security.Steganography and steganalysis are mutually antagonistic and interdependent.In order to protect the information security in all aspects,it is urgent to research and innovate the techniques for both steganography and steganalysis.The traditional steganography and steganalysis mainly focus on the carrier of digital images,while the research of steganography and steganalysis based on digital videos is still in its infancy.In today's digital era,digital video is characterized by wide source,large capacity and strong camouflage.It is not only an excellent carrier for steganography,but also a key object need to be prevented in steganalysis.Therefore,studying the steganography and steganalysis based on digital videos meets the needs of information security,and is also of great academic meaning and application value.The complexity of digital video in structure and coding brings many challenges to the research and application of video steganography and steganalysis.For video steganography,the digital video has various types of embedding domains;however,existing video steganographic methods only use a single embedding domain to hide secret messages,and cannot fully utilize the embedding space in videos.For video steganalysis,the complicated video coding process,inadequate statistical features and limited prior knowledge of embedding domains also pose difficulties to the practical steganalytic detection.This dissertation,aiming at the development and application requirements of video steganography and steganalysis,focuses on the problems of video steganographic security and steganalytic feature construction,and studies emphatically on the multi-domain embedding strategies for video steganography,the feature design in motion vector domain steganalysis and the universal detection of video steganography in multiple domains.Finally,the dissertation proposes a series of key technologies and methods for video steganalysis and steganalysis.The main work includes the following four aspects.1)Video steganography based on multiple embedding domains.Aiming at the problem that multiple embedding domains in video carrier cannot be used at the same time,this dissertation proposes a multi-domain video steganography by combining partition modes(PMs)and motion vectors(MVs).To prevent the mutual interference between multiple embedding domains,two multi-domain embedding strategies are designed,namely sequential embedding strategy and simultaneous embedding strategy.The sequential embedding strategy respectively embeds the message into PM domain and MV domain in two stages,according to the unidirectional relationship between the two domains.The simultaneous embedding strategy is accomplished in one stage by embedding domain mapping and embedding cost adjustment,which make the two associated embedding domains independent of each other.Both multi-domain embedding strategies can eliminate the interaction between different embedding domains,and also improve the steganographic security and increase the steganographic capacity.2)Video steganalysis based on generalized local optimality in motion vector domain.Aiming at the application problem of video encoding characteristics in steganalysis,this dissertation proposes to construct steganalytic features by using generalized local optimality in MV domain.The steganographic embedding in MV domain destroys the local optimality of MVs,so the accurate estimation of the local optimality is vital for feature design.This dissertation generalizes the traditional local optimality in two aspects.First,according to the interaction of the embedding changes on adjacent MVs,the local optimality estimation of the MVs is extended from the fixed predicted motion vectors(PMVs)to the dynamic ones.Second,in light of the modification pattern of PMVs,the local optimality of MVs is extended to the local optimality of the PMVs.The steganalytic features are extracted based on the two types of generalized local optimality,which can precisely capture the statistical changes caused by steganographic embedding,increase the statistical diversity,and finally significantly improve the detection accuracy.3)Video steganalysis based on combined and calibrated features in motion vector domain.Aiming at the problem that the statistical information used for video steganalytic features is not rich enough,this dissertation combines the neighborhood optimality of MVs and the distributions of motion vector differences(MVDs)to construct steganalytic features,which are further enhanced by calibration.First,to decrease the negative effect of quantization distortion on coding optimality,PMs are used to measure the quantization distortion.Moreover,the steganographic embedding changes not only the local optimality of MVs,but also the neighborhood optimality of MVs.By integrating these two aspects,the partition based neighborhood-optimal probability features are extracted.Second,the MVDs are used to replace the traditional neighboring motion vector differences,and the co-occurrence matrix features are constructed based on the inter and intra statistical characteristics of MVDs.Third,a window-optimal calibration is designed by using the optimality of MVs in a window area to recover original MVs.The final combined and calibrated features have strong detection capabilities and can be applied to various video coding conditions.4)Multi-domain video steganalysis based on the consistency of motion vectors.Aiming at the problem that multiple embedding domains in videos are difficult to universally detected,this dissertation proposes a universal video steganalytic method for PM domain and MV domain.The MVs of sub-blocks in the same macroblock usually have different values,and the steganographic embedding in either PM domain or MV domain can make the MVs of the sub-blocks tend to be consistent in values.According to this phenomenon,this dissertation presents the concept of motion vector consistency,and constructs a universal steganalytic feature set based on motion vector consistency.The feature set can simultaneously detect the video steganography in PM domain and MV domain,and also has the characteristics of high detection accuracy,strong robustness,and low complexity.