An Extendible Hashing Structure for Image Similarity Searches

  • Author(s) / Creator(s)
  • Technical report TR00-04. Image similarity search is employed to retrieve similar images from a database. Colour features, particularly colour histograms, are extensively utilized to compute the image similarity. This thesis presents a multi-scale representation of image colour histograms that allows queries at different precision levels as well as queries on part of an image instead of an entire image. % (57) Because image databases usually store a large number of images, sequential scanning of the database is not feasible; an index scheme is needed to speed up the similarity search. A new index structure --- three-dimensional extendible hash --- is designed to index image average colours. The average colours are used in a two-step colour similarity search: the index is searched first to filter the database before colour histogram comparisons are applied. Experiments have been made to test the performance of the three-dimensional extendible hash, and the results are discussed in this thesis. | TRID-ID TR00-04

  • Date created
  • Subjects / Keywords
  • Type of Item
  • DOI
  • License
    Attribution 3.0 International