Research On Digital Video Temporal Forgery Detection Based On Structural Similarity

As a carrier of information, digital video has been widely used in many fields, such as news reports, commercial propaganda etc. Some adverse effects will appear in many aspects of political and social, if an important video which has been malicious attacked is spreading on the Internet. In response to the actual needs, increasingly scholars enter into the field of digital multimedia forensics. With further research, those techniques seek to provide information about digital multimedia content without relying on external descriptors or extrinsically implanted information. Instead, these techniques make use of fingerprints left in digital content by editing operations or the digital capture process.Existing forensics technologies mainly based on the traces left by the video editing and video capture device. Scholars explore and make use of them for video tampering detection. The achievements include extraction the fingerprints left by video software, searching the origin of video, tracking the process of digital multimedia content. However, there are some problems are still exiting, for instance, high dimensionality features, computational complexity, the limitations of practical application etc. In this paper, aiming at the temporal forensics, our main work includes the following two aspects:(1) By removing some frames which contain a crime scene or crime evidence, the forger can change the video content. A new method based on structure similarity is proposed for frame remove forgery in this paper. To facilitate the transmission and storage, most videos accept lossy compression basically, which leading to similarity between frames decreased with the increase of distance. According to this feature, the proposed method employs the structure similarity to measure the similarity between adjacent frames, and uses the threshold as evidence to find out outliers, and achieves frame deletion forgery detection and localization. Experiment results demonstrate that the proposed method can localize the point where frames has been deleted for videos captured by stationary camera. Even if the number of frames has been deleted is an integer multiple of the GOP length of the tampered video, the algorithm can work and not affected by recompression, format conversion.(2)Frame duplication forgery is a very common operation for video tampering in the temporal domain. A lot of works have been proposed for detecting this type of tempering operation. However, there are two disadvantages existing. The first one is using fixed threshold; another is huge computation. In general, frame duplication forgery and re-compression would be performed on a video at the same time. Since re-compression causes data lost, the fixed thresholds may lose their effects. Therefore, an algorithm based on dynamic threshold is proposed in this paper, which can improve the robustness. Additionally, the proposed method adopts dictionary order algorithm to reduce the computation. Three performance indices:precision, recall and computer time are employed to evaluate our algorithm. The results demonstrate that the proposed method outperform the existing methods in terms of precision, recall and computation time.