An Efficient Image Similarity Measure Based on Approximations of KL-Divergence Between Two Gaussian Mixtures

  • Authors:
  • Jacob Goldberger;Shiri Gordon;Hayit Greenspan

  • Affiliations:
  • -;-;-

  • Venue:
  • ICCV '03 Proceedings of the Ninth IEEE International Conference on Computer Vision - Volume 2
  • Year:
  • 2003

Quantified Score

Hi-index 0.12

Visualization

Abstract

In this work we present two new methods for approximating theKullback-Liebler (KL) divergence between two mixtures of Gaussians.The first method is based on matching between the Gaussian elementsof the two Gaussian mixture densities. The second method is basedon the unscented transform. The proposed methods are utilized forimage retrieval tasks. Continuous probabilistic image modelingbased on mixtures of Gaussians together with KL measure for imagesimilarity, can be used for image retrieval tasks with remarkableperformance. The efficiency and the performance of the KLapproximation methods proposed are demonstrated on both simulateddata and real image datasets. The experimental results indicatethat our proposed approximations outperform previously suggestedmethods.