Monitoring near duplicates over video streams

  • Authors:
  • Xiangmin Zhou;Lei Chen

  • Affiliations:
  • CSIRO, Canberra, Australia;Hong Kong University of Science and Technology, Hong Kong, China

  • Venue:
  • Proceedings of the international conference on Multimedia
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Since near duplicates are ubiquitous over different data sources, increasing research efforts have been put to near duplicate detection recently. Among all the near duplicate detection tasks, an important one is continuous near duplicate monitoring over video streams. Existing video monitoring techniques are not effective for handling the variations that commonly exist among near duplicates. Moreover, approaches proposed for the near duplicate detection in archived video databases are inefficient when applied to high speed video streams. In this work, we propose a framework for effectively online monitoring near duplicates over video streams. Specifically, we first propose a novel representation, a video cuboid signature, to describe a video segment. To capture the local spatio-temporal information of video subclips, we employ the Earth Mover's Distance (EMD) to measure the similarity between two signatures. Both the signature construction and the sequence similarity measure are incrementally processed by exploiting the inherent property of signature series. Then, we propose a novel scheme called locality sensitive multi-leveled approximation (LSMA) that optimizes the near duplicate video similarity matching over streams based on the locality sensitive hashing under EMD metric. The extensive experiments demonstrate the high performance of our approach in terms of the detection accuracy and time cost.