Handle local optimum traps in CBIR systems

  • Authors:
  • Danzhou Liu;Kien A. Hua;Hao Cheng

  • Affiliations:
  • University of Central Florida, Orlando, Florida;University of Central Florida, Orlando, Florida;University of Central Florida, Orlando, Florida

  • Venue:
  • Proceedings of the 2008 ACM symposium on Applied computing
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Existing CBIR systems, designed around query refinement based on relevance feedback, suffer from local optimum traps. That is, when the user is examining a relevant cluster surrounded by less relevant images, essentially the same set of images will be returned for the user to provide relevance feedback. Since the user would select the same query images again, the relevance feedback process gets trapped in a local optimum. This local-optimum trap problem may severely impair the overall retrieval performance of today's CBIR systems. In this paper, we therefore propose a simulated annealing-based approach to address this important issue. When a stuck-at-a-local-optimum occurs, we employ a neighborhood search technique (i.e., simulated annealing) to escape from the local optimum. We also propose an index structure to speed up such neighborhood search. Our experimental study confirms that our approach can efficiently address the local-optimum trap problem, and therefore can improve the effectiveness of existing CBIR systems.