Shape transformation-based image retrieval using non-uniform rational B-spline
Loading...
Date
2003-06
Authors
Liang, Kim Meng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Large image databases are used in many multimedia applications such as
entertainment, business, art, engineering and science. Searching information in these
databases is a crucial problem to be solved for the development of visual information
system. Therefore efficient retrieval methods are required for the purpose of finding a
desired image from a collection of images.
This research is focussed on developing a shape-based image retrieval system for use
in thematic databases. In thematic databases, all the images are highly similar to each
other. For this purpose, a novel shape similarity matching methodology is needed in
order to detect a small similarity difference between two images. This similarity
difference is indicated by the computed similarity measure that is used to display a set
of relevant images from the databases. The similarity measure introduced here bears
resemblance to the human notion of'similarity. In the real-time shape-based image
retrieval systems, the accuracy of the retrieval results and the retrieval time are two
aspects that need to be looked into before any application is possible.
In the past decade, shape similarity matching methodology has evolved and is
influenced by two general approaches: feature vector approach and shape
transformation approach. Feature vector approach provides fast retrieval time but does
not ensure high accuracy retrieval result, whereas shape transformation approach has
the ability to ensure high retrieval accuracy but it involves high computation time.
Thus a new methodology, referred to as NURBS-warping approach is implemented by
integrating the advantages of the two approaches. In the proposed approach, NonUniform
Rational B-Spline (NURBS) and Gradient Vector Flow (GVF) are
incorporated fo ensure accurate and fast retrieval results. NURBS is a compact and
accurate shape descriptor, whereas GVF ensures fast and accurate matching results.
The effectiveness of the NURBS-warping approach is examined by carrying out
experiments on a collection of one thousand and one hundred highly similar images
from a thematic fish database. The overall retrieval results show that the proposed
approach is able to derive an accurate similarity measure that resembles human
similarity judgement.
Description
Keywords
Efficient retrieval methods are required for the purpose of , finding a desired image from a collection of images