Back to search

University of Southampton

Content based image retrieval : analogies with text

Abstract

dc:description.abstract

<p>This thesis describes research into an area of content based image retrieval (CBIR), that of feature indexing for the purpose of rapid retrieval. The techniques in this thesis draw from the field of text IR and demonstrate that individual image extraction algorithms can be optimised for use with an inverted index, which could lead to CBIR systems capable of sub-second retrieval times on collections of millions of images.</p> <p>A novel global feature algorithm, QMNS, is presented, which is capable of capturing both colour and texture information in a spatially insensitive manner. Images are divided into regular patches from which dominant colour modes are derived using the mean shift algorithm. The RGB bi-modal colour space is quantised giving a set of labelled feature terms with associated frequencies and the terms inserted into an inverted index. Terms in the index are retrieved with a TF*IDF algorithm.</p> <p>The performance of QMNS and the index is measured by comparison with an RGB colour histogram, an RGB CCV histogram, and the unquantised MNS features. Precision and recall results show that the indexed feature performs equally as well as the other algorithms for an image collection of photographic images. An analysis of the distribution of each type of feature term was performed, showing that Zipf 's law holds in each case. Quantisation parameters for the algorithms were varied, demonstrating that a tradeoff exists between vocabulary size, the average precision, and the speed of retrieval.</p> <p>This thesis indicates that the current generation of highly successfully text IR systems, which are implemented using inverted indexes, could provide the basis for very rapid image and multimedia retrieval. The optimisation techniques shown can be used for any quantisable feature, and allow retrieval to be performed without specialised comparison algorithms.</p>

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2005

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Westmacott, Mike

Rights

Language dc:language
English, eng

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Westmacott, Mike. Content based image retrieval : analogies with text. doctoral thesis, University of Southampton, 2005.