| As an effective method of covert communication for messages of large volume, video steganography has attracted much attention. To effectively detect harmful covert communication activity based on video steganography, video steganalysis has important researching senses and application prospects. This paper mainly focuses on video steganalysis. To detect steganography on several data domains, corresponding steganalytic methods are proposed.In addition, domain adaptation is combined with universal steganalytic methods in this paper, and the applicability is analyzed on solving the problem of cover mismatch.The main works and innovations are as follows.1.Steganalysis on compressed videos against spread spectrum steganography. A novel steganalytic feature named SPEAM is proposed, which combines spatial redundancy and temporal redundancy of the video content. When subjected to uncompressed videos, the performance of SPEAM feature is similar to state-of-art image steganalytic feature and is much better than existing video steganalytic features. The advantange of the proposed feature is that it performs better than existing image steganalytic features and video steganalytic features on compressed videos. Furthermore, the computation complexity of SPEAM feature is quite low.2. Steganalysis against motion vector (MV) based steanography. Existing MV based steganalytic methods can be classified into two categories. However, the first category performs unfavorable and not universal enough, and the other category suffers from the difficulty of ensuring re-encoding parameters to be same as primitive encoding parameters. Based on the assumption that MVs are local optimal, a novel steganalytic feature named AoSO is proposed. AoSO feature performs significantly better than existing two categories, and is more universal. Specifically, it is applicable for various cases of coding parameters.3.Steganalysis against H.264 intra prediction mode (IPM) based steganography. According to the H.264 standard, a calibration schema of IPM is presented, which utilizes SATD (Sum of Absolute Transformed Difference) as the encoding cost. The further steganalytic feature named CIPM is proposed, which is of low complexity and is significantly more favorable than existing features.4.Improving the performance of universal steganalytic features for cover mismatch. A feature based domain adaptation method, named TCA, is studied. MV based steganalysis and TCA are combined for solving the problem of steganography methods mismatch. Experiments are carried out to validate the potential improvements introduced by TCA. Based on issues occurred in those experiments, the capability of domain adaptation methods like TCA on steganalysis is discussed, and the localization of domain adaptation method on steganalysis is pointed out.5.Steganalysis against the spatial steganography tool named MSU StegoVideo.Existing steganalytic methods against MSU Stego Video are reviewed, which suggests that those methods mainly focus on the detection of hidden message and are not favorable enough. Three types of fingerprints introduced by MSU Stego Video are analyzed, including the specific colorspace transformation, chessboard mask pattern, and the controlling mechanism of the embedding intensity. Based on those fingerprints, a novel method, named MP algorithm, is proposed for both detecting the hidden messages and estimating the embedding parameter "noise level".MP algorithm performs better than existing methods on the detection accuracy, and can further roughly estimate "noise level" and the stego information sequence.6.Steganalysis against the format based steganography tool named OpenPuff. According to specifications of corresponding video containers and the analysis on distinctions introduced by steganography, corresponding steganalytic methods, including methods for blind detecting and extracting hidden messages, are proposed for detecting and extracting hidden messages embedded in videos of MPEQ VOB, FLV, MP4, and 3GP formats.
|
PDF Full Text Request