Efficient benchmarking of content-based image retrieval via resampling

Jialie Shen*, John Shepherd

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

7 Citations (Scopus)

Abstract

While content-based image retrieval (CBIR) is an expanding field, and new approaches to ever more effective retrieval are frequently proposed, relatively little attention has so far been paid to the process of evaluating the effectiveness of CBIR methods. Most of the reported evaluations use standard IR evaluation methodologies, with little consideration of their statistical significance or appropriateness for CBIR, which makes it difficult to assess the precise impact of individual methods. In this paper, we present a new approach for evaluating CBIR systems which provides both efficient and statistically-sound performance evaluation. The approach is based on stratified sampling, and provides a significant improvement over existing evaluation approaches. Comprehensive experiments using our approach to evaluate a range of CBIR methods have shown that the approach reduces not only the estimation error, but also reduces the size of the test data set required to achieve specific estimation error levels.

Original languageEnglish
Title of host publicationProceedings of the 14th Annual ACM International Conference on Multimedia, MM 2006
Pages569-578
Number of pages10
DOIs
Publication statusPublished - 01 Dec 2006
Externally publishedYes
Event14th Annual ACM International Conference on Multimedia, MM 2006 - Santa Barbara, CA, United States
Duration: 23 Oct 200627 Oct 2006

Publication series

NameProceedings of the 14th Annual ACM International Conference on Multimedia, MM 2006

Conference

Conference14th Annual ACM International Conference on Multimedia, MM 2006
Country/TerritoryUnited States
CitySanta Barbara, CA
Period23/10/200627/10/2006

Keywords

  • Evaluation
  • Image retrieval
  • Sampling

ASJC Scopus subject areas

  • General Computer Science
  • Media Technology

Fingerprint

Dive into the research topics of 'Efficient benchmarking of content-based image retrieval via resampling'. Together they form a unique fingerprint.

Cite this