Research On Feature Extraction Algorithm For Video Tamper Detection

Video, with its convenient, intuitive and rich information content becomes the primary network information bearing forms. However, the open Internet environment is bound to have a large number of potential video security risks. Video of malicious tampering has become the main form of video attacks, provides opportunities for criminals to spread bad information. In addition, video from a single form of tampering with the frame insert and add to the current combination of a variety of tampering. These tampered video is not only change the video content, but also introduces the interference of noise, which increased the difficulty of the tamper detection. Based on the above issues, the main purpose of this paper is to study the active tamper detection, by extracting fingerprint features of tampering video and comparison with the source video fingerprint features, to determine whether the video be detected and tracing the tamper area.In this paper, based on the video tampering localization as the breakthrough point, study on video tamper detection. We analyzed the problems existing in the existing algorithms. We combining a large number of the original video features, construct the fusion of global features and local features of the algorithm, and step by step from coarse to fine analysis of the video tamper detection and accurate positioning.First, in order to detect the tampered video, we put forward a method of Zernike feature. According to the characteristics of the video tamper form(color of tampering, scale of tampering, crop of tampering, video frame contents of tampering). We used the low Zernike moment, which has rotation invariance and well describe the image shape feature, to realize the tamper detection. At the same time, we integrate into the color component when extracting Zernike moment to detect the video tampering of color.Secondly, Zernike feature detection only achieve the overall video frame tamper detection, it can’t locate the tampered area. So, we proposed ORB features to achieve accurate positioning based on the analysis of video tampering. ORB feature can effectively describe the local information of image, but it exists interference noise when locate the tamper area. In this paper, we put forward the energy minimization principle and combined with sparse characteristics of noise features to remove the surrounding noise points. The experimental results show that this method can effectively detect the video tamper and locate tamper areas.Finally, from the view of the application of video surveillance certification, we built a video tamper detection system based on embedded Tiny210 development platform. In the development of the board, we use Tiny210 to capture video, extract fingerprint, encode and decode video, and also transmit network data. In the client, the MFC is used to realize the video receiving and playing, the video fingerprint receiving and storing, the off-line video fingerprint extraction and the video tamper detection. We design the multithreaded Linux program to realize video capture, fingerprint data extraction and RTP/RTCP to transmit video. And also, we put forward the realization of the tampered video detection method, analyzing the system performance and security.