A Video Copy Detection System Based On Graphs

As the Internet and digital video technology develop, a large number of video content generated by users from video sharing websites set up. The user can upload modified video to the video websites, and a lot of repetitions of video production come out. It produced an urgent need to the digital video retrieval, management and copyright protection. Video copy detection based content (shortly referred to as video copy detection) takes an important role in video information management, filter and copyright protection. Through effective video copy detection technology, we can quickly find out copies of the digital video. The technology can be used to effective protection of digital videos, and also can be used to monitor the spread of illegal video. In recent years, video copy detection becomes a new frontier research field.In the video copy detection technology, two problems need to be solved. One is accuracy. It need determine the granularity of the detection, according to specific application. The other is efficiency. Face to a big video database, this is a typical data intensive and computing intensive problem. This paper will detect the internet videos which add LOGO, subtitles and are changed some resolution, and dive into the study of efficiency. In recent two years, a popular distributed computing model, MapReduce model, is used to the application of large-scale video copy detection operations. With MapReduce model, we design and implement of distributed video copy detection system prototype. The paper mainly works are as follows:First, video copy detection algorithm used in our paper is proposed. Because the technology of video copy detection has diversity, we focused on specific application adopted in the system of the algorithm. It contains key frame extraction strategy, which is suitable for us to use this system, feature calculation, and how to calculate video similar degrees.Second, video copy detection algorithm based MapReduce distributed computing model is proposed. MapReduce is an effective solution to solve the data intensive and computing intensive tasks. According to the nature of the video copy detection, this paper proposes a MapReduce implementation and a prototype system, which is used to verify the algorithm.Third, we design experiments in speed-up, scale-up and size-up, to verify our algorithm and the realization of the system. The experimental results show that the algorithm can effectively improve the efficiency of the video copy detection. The paper also discusses the index of the video. The algorithm in the pater is based on matching algorithm of brute force, and we hope a follow-up study into the video copy detection with index structure.This paper connects the field of hot issues in video research -- video copy detection and distributed computing hotspot -- MapReduce. The author puts forward his own algorithm and the design of prototype system, expands the distributed computing model MapReduce in video processing field and video information retrieval technology application. It is also a meaningful try.