A note on the application of Web information to near-duplicate online video detection

摘要

This paper presents a novel approach for detecting near-duplicates in online video search results. A vast majority of existing approaches focuses only on the low-level image information. This can be inefficient when the size of the video collection is large. We propose to utilize existing Web information, i.e. the video description, tags and comments to determine relationships within the search results and obtain an initial indication of the near-duplicates prior to applying computationally expensive conventional approaches. We compare the effectiveness of conventional approaches and Web information-only approaches, and investigate ways of combining the two approaches to improve the effectiveness of near-duplicate video detection.