Clip-based hierarchical representation for near-duplicate video detection

Searching for near-duplicate content has become an important task in many multimedia applications, for example, images, videos and music. The ability to detect duplicate videos plays an important role in several video applications, for example, effective video search, copyright infringement and the study on users’ behaviour on near-duplicate video production. Current web video search systems rely only on text keywords and, hence, fail to detect many duplicate videos. In this paper, we analyse the problem of near-duplicate detection and propose a practical solution for real-time large-scale video retrieval. Unlike many existing approaches which make use of video frames or key-frames, our solution is based on a more discriminative signature of video clips. The feature used in this paper is an extension of ordinal measures which have proven to be robust to change in brightness, compression formats and compression ratios. For efficient retrieval, we propose to use multi-probe locality sensitive hashing (MPLSH) to index the video clips for fast similarity search and high recall. MPLSH is able to filter out a large number of dissimilar clips from video database. To refine the search process, we apply a similarity voting based on video clip signatures. Experimental results on the dataset of 12,790 web videos show that the proposed approach improves average precision over the baseline colour histogram approach while satisfying real-time requirements.