Video Data Feature Reduction And Its Application In Tamper Detection

Many places have already set up the monitoring equipment at present, in order to obtain and save the important video data, which aimed at ensuring security. If someone does the malicious tampering on purpose to achieve the bad intention, it will result in the negative influence to a great extent. Therefore, detecting the circumstance of these essential video data has the considerable practical significance. To solve this problem, the study of video tamper detection combines various knowledge of professional field, and get more and more attention at this stage. In this paper, we first consider extracting the different characteristics of video data, then reduce the dimension of it and compress, finally detect and identify the tampering circumstance of video data using it.We utilize the related methods of the feature reduction, and apply it to the video tamper detection, do the following work:(1) A novel algorithm which solve how to treat the non-uniform distribution data and how to use the distant points is proposed in the field of the feature reduction, named Harmonic Mean Geodetic-Kernel LLE(HMG-KLLE) algorithm; (2) By leading into the compressed sensing theory, in order to diminish the redundancy rate of the video noise characteristic information, we also propose a video tamper detection algorithm based on compressed sensing and has a good detection result; (3) On the basis of (1), we reduce the dimension of the video textual feature fusion information, develop a new video tamper detection algorithm based on manifold learning and gain a better detection result.The experimental results show we extract the different video characteristics and apply it to the field of video tamper detection, can get a quite well tamper identification rate.