| In internet, there are a lot of duplicated videos. How to efficiently detect near duplicate in large scale videos has been a hot topic for a long time. A high efficient near duplicate detection algorithm is useful in many ways. Firstly, video copy detection can be used to calculate the similarity of videos. So it can be used to content based video search missions. Currently, when customers search videos, mostly the search is based on video title or author information. Video copy detection can be used to search videos by the content of this video. It’s convenient to search appropriate videos resources. Secondly, video copy detection can be used to eliminate duplicate videos to improve quality of the search result. When customers search a video, they always hope the duplicate result the less the better. Copy detection can provide more valuable information by eliminate duplicate information. Thirdly, video copy detection can be used to detect illegal videos. It’s important to protect copy-right. Lastly, many other missions are based on video efficient copy detection, such as video cluster mission and so on. One of the most important steps of those mission is calculate the similarity of two videos. Research of how to improve the performance of big scale video copy detection is important to all of those missions.In recent years, serial video copy detection algorithm can’t be used to the increasing video set. In this paper we use GPU parallel computing to parallelize this video copy detection mission. In parallel environment, three problems should be considered to solve the parallel video copy detection problem. First is to extract the video features which are suitable to parallel computing. Second is to design a parallel video copy detection algorithm. Third is to optimize the algorithm based on the property of GPU hardware. Former research failed to handle these three problems. In this paper, an efficient video copy detection algorithm based on GPU is raised. |
PDF Full Text Request