Research On Steganalysis Against Motion Vrector-based Video Steganography

Steganography is the art of hiding the secret message in a variety of popular media data on the internet, such as audio, image and video. A safe path is built between the sender and the receiver of communication, and it is difficult for interceptor to detect its existence. Once the terrorist use steganography to do secret communication, like planning terrorist attacks, it will certainly be very dangerous. So research on methods of steganalysis to attack steganography becomes very imminent. How to detect the media files such as pictures, videos, web pages whether hidden information, has become an important research question in the field of information security.Steganography and steganalysis in images and raw video have wide literature now. Video is composed by a series of images, which have high correlation in spatial and temporal domain. Compared to single image, video have greater volume, it is possible to accommodate more secret message. Coupled with the video in the network is very popular, so that messages embedded in video can be spread more quickly and easily. These factors are described above show that steganography based on compressed video has a good environment to develop:effective mechanism and rich carrier. Motion vector in compressed video is lossless coded and transmitted, so it make motion vector the perfect carrier for steganography. Thus, therefore it becomes more significant to pay attention to video steganalysis against MV-based steganography.Based on the review of video coding theory, video steganography and steganalysis, we create a system model and analyze the characteristics of the model to research the video steganalysis method. The main research work and innovations are as follows:1. Proposed steganalysis method based on spatial and temporal correlation of motion vector. Motion vectors only exist in video, so that image steganalysis methods cannot be used to detect steganographic methods based on modifying motion vector. Motion vectors represent the movement of objects, they have strong spatial and temporal correlation, but steganography destroy this correlation, experiments show that it can greatly improve the detection accuracy by extracting temporal correlation Markov feature. 2. Proposed steganalysis features based on prediction error. After modifying motion vectors in video steganography, the prediction error will be recalculated and further be coded, so the visual impact of the video quality is very small. It almost impossible to catch differences just by human visual perception, furthermore the image steganalysis working in pixel domain will fail, too. But it also brings a disadvantage that modification of the motion vector make prediction error larger. Experiments show that residuals histogram features can further improve the accuracy of detection.3. Proposed a steganalysis method based on COST sorting of re-encoded video. Steganographic method modifies the motion vector, so it makes the prediction error larger, because the motion vector is no longer optimal after modifying. While reencode the video, motion vector have a trend to recover from non-optimal. After sorting the motion vector’s COST, then construct25-dimensional feature capture the location of motion vector in sorting list. Experiments show that features extracted from re-encode video can further perform greater than other comparing features.