Mean Shift Based Clustering in High Dimensions: A Texture Classification Example

  • Authors:
  • Bogdan Georgescu;Ilan Shimshoni;Peter Meer

  • Affiliations:
  • -;-;-

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

Quantified Score

Hi-index 0.00

Visualization

Abstract

Feature space analysis is the main module in many computer visiontasks. The most popular technique, k-means clustering, however, hastwo inherent limitations: the clusters are constrained to bespherically symmetric and their number has to be known a priori. Innonparametric clustering methods, like the one based on mean shift,these limitations are eliminated but the amount of computationbecomes prohibitively large as the dimension of the spaceincreases. We exploit a recently proposed approximation technique,locality-sensitive hashing (LSH), to reduce the computationalcomplexity of adaptive mean shift. In our implementation of LSH theoptimal parameters of the data structure are determined by a pilotlearning procedure, and the partitions are data driven. As anapplication, the performance of mode and k-means based textons arecompared in a texture classification study.