Contents:
- 1 Introduction
- 2 Related work
- 2.1 Concept-based video ranking functions
- 2.2 Uncertainty in text retrieval
- 3 The uncertain representation ranking framework
- 4 Shot retrieval
- 5 Segment retrieval
- 6 Experiments
- 7 Discussion
- 8 Conclusions
- References
As concept based video retrieval often relies on imperfect and uncertain concept detectors, the researchers propose in this paper a general ranking framework to define effective and robust ranking functions through explicitly addressing detector uncertainty. The uncertain representation ranking framework (URR framework) considers basic ranking functions adapted from text retrieval based on representations of known concept occurrences. The experiments (section 6) with which the performance of the URR framework is evaluated were carried out on five TRECVid collections and two collections which use simulated detectors of varying performance. They proved the general framework’s ability to produce effective and robust ranking functions by applying it to two retrieval tasks: shot retrieval and (long) segment retrieval. Not only do the results show a significant improvement over most other retrieval methods from other uncertainty classes; both methods also continue to deliver a strong performance compared to other methods as the concept detector performance improves. Part of the work reported here was funded by the EU Project AXES (FP7-269980). This article appeared in Information Retrieval, October 2013, Volume 16, Issue 5, pp 557-583.
This research paper provides those responsible for AV archive cataloguing and access with a good impression on what is possible nowadays in the field of concept searching and relevance ranking based on automatic indexing of shots and segments.