Content-Based Video And Audio Copy Detection

With the rapid development of Internet technology, digital media comes to be more and more popular. As a result, multi-media has become the main information transported via the internet and the most important issue becomes how to retrieve efficiency in the branch of media data sets. Content-based video and audio copy detection is a new focus in video retrieval which can resolve the problem by describing the video content in a bottom-up way. As one special application of CB VR, video and audio copy detection (CBCD) attracts more and more attention. The main application of CBCD is video and audio copyright protection and content monitoring.This paper mainly focuses on two issues:the efficiency and precision of video and audio copy detection. And two detection systems are developed too. The details are as follows:In video copy detection, we trained a 50,000-dimension codebook using parallel computing method so as to obtain much higher precision and efficiency. And in order to make the detection more efficient, a two-level codebook algorithm was proposed. At last, a data storing structure with inverted index was presented and implemented based on the visual database index so that the speed of accessing database was increased.In audio copy detection, two frameworks were proposed according to different audio fingerprints. First, we found out a framework based on traditional MFCC audio feature and Hilbert space-filling curve indexing. And then, another framework called energy different fingerprint was also proposed and a hash indexing system was set up for it. After comparison between the two frameworks in performance, the latter one was selected for our audio detection.And finally, this paper merged the results gained from the two video-audio systems and thus attended the TRECVID 2010. In the competition, the performance of the system was better than the average so as to prove a high efficiency.