Commercial film detection and identification based on a dual-stage temporal recurrence hashing algorithm

  • Authors:

  • Xiaomeng Wu;Narongsak Putpuek;Shin'ichi Satoh

  • Affiliations:

  • National Institute of Informatics, Tokyo, Japan;Chulalongkorn University, Bangkok, Thailand;National Institute of Informatics, Tokyo, Japan

  • Venue:
  • Proceedings of the international workshop on Very-large-scale multimedia corpus, mining and retrieval
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposes a dual-stage temporal recurrence hashing algorithm for fully unsupervised and super-fast Commercial Film (CF) mining in large-scale broadcast video archives. The first-stage hashing algorithm converts a large amount of video segments into a set of temporal occurrence patterns on the basis of a luminance-based fingerprinting strategy. During the second stage, each temporal recurrence of a certain segment is mapped to an inverted index to build a recurrence hashing histogram, which is the key idea in this paper to achieve super-fast CF detection and identification. The detection and identification task is then converted into one of detecting local maximums from this hashing histogram, with one local maximum corresponding to one pair of two identical CF segments. A large-scale archive, containing a 10-hour and a 1-month sequence, was used for experimentation. Our algorithm performed CF detection and identification on the 1-month sequence in merely 87 minutes and had 90.59% detection accuracy and 98.06% localization accuracy.