A New Indexing Scheme for Content-Based Image Retrieval

  • Authors:

  • Guang-Ho Cha;Chin-Wan Chung

  • Affiliations:

  • Dept. of Information and Communication Engineering, Korea Advanced Institute of Science and Technology, 207-43 Cheongryngri, Dongdaemun, Seoul 130-012, South Korea. E-mail: ghcha@tmic.tit.ac.kr;Dept. of Computer Science, Korea Advanced Institute of Science and Technology, 373-1 Kusong-Dong Yosong-Gu, Taejon 305-701, South Korea. E-mail: chungcw@mgis.kaist.ac.kr

  • Venue:
  • Multimedia Tools and Applications
  • Year:
  • 1998

Quantified Score

Hi-index 0.00

Visualization

Abstract

We propose a new efficient indexing scheme, calledthe HG-tree, to support content-based retrieval in imagedatabases. Image content is represented by a point in amultidimensional feature space. The types of queries consideredare the range query and the nearest-neighbor query, both in amultidimensional space. Our goals are twofold: increasing thestorage utilization and decreasing the area covered by thedirectory regions of the index tree. The high storageutilization and the small directory area reduce the number ofnodes that have to be touched during the query processing. Thefirst goal is achieved by suppressing node splitting ifpossible, and when splitting is necessary, converting two nodesinto three. This is done by proposing a good ordering on thedirectory nodes. The second goal is achieved by maintaining thearea occupied by the directory region as small as possible. Thisis done by introducing the smallest interval that encloses allregions of the lower nodes. We note that there is a trade-offbetween the two design goals, but the HG-tree is so flexiblethat it can control the trade-off to some extent. We present thedesign of our indexing scheme and associated algorithms. Inaddition, we report the results of a series of tests, comparingthe proposed index tree with the buddy-tree, which is one of themost successful point indexing schemes for a multidimensionalspace. The results show the superiority of our method.