City University of Hong Kong

CityU Institutional Repository >
3_CityU Electronic Theses and Dissertations >
ETD - Dept. of Electronic Engineering  >
EE - Doctor of Philosophy  >

Please use this identifier to cite or link to this item:

Title: Evolutionary algorithm-based affine-invariant matching of object shapes from broken boundaries
Other Titles: Ji yu jin hua suan fa zhi po sui wu jian lun kuo xian fang she bu bian xing pi pei
Authors: Situ, Wuchao ( 司徒武超)
Department: Department of Electronic Engineering
Degree: Doctor of Philosophy
Issue Date: 2011
Publisher: City University of Hong Kong
Subjects: Optical pattern recognition.
Image processing -- Digital techniques.
Computer vision.
Notes: CityU Call Number: TA1650 .S57 2011
xii, 107 leaves : ill. 30 cm.
Thesis (Ph.D.)--City University of Hong Kong, 2011.
Includes bibliographical references (leaves [94]-105)
Type: thesis
Abstract: Viewpoint invariant matching of object shapes is a fundamental but difficult problem in computer vision. It pertains to the recognition or alignment of shapes that are subject to certain geometric transformations caused by different viewing positions. In practice, the images of objects captured by optical means are affected by the illumination, as well as various kinds of artifacts; the objects often present fragmented, and more severely, incomplete contours in shape. This imposes further complications on the problem. This thesis reports the developments of several shape matching schemes, which can work independently or corporately in identifying object images that are captured under poor lighting conditions. Specifically, the proposed methods can be applied in the matching of near-planar object shapes from broken boundaries, and moreover, under noise contamination. One of the main emphases of my works, is to maintain a high success rate in shape matching, and at the same time minimizing the computation time. For near-planar objects, the matching process can be posed as an optimization problem in realising an affine transform that yields a best matching score between a pair of contours. Along this direction, simple genetic algorithms (SGA) and particle swarm optimization (PSO) have been proven effective. Despite the moderate successes of these approaches, however, they present erratic performance amongst different objects, with reduced success rates and prolonged computation times in some cases. These shortcomings can be attributed to the lack of an initial population/swarm community that can realise global solutions. In this thesis, a solution to this problem is presented by integrating PSO with the migrant principle (MP). The latter is analogous to immigrant policy in real-world situations; it introduces a continuous influx of foreign candidates to the swarm community to promote diversity, and therefore, exploration power in the population. Evaluations show that the method is less sensitive to swarm size and can achieve high success rates for all test samples based on a small swarm community. To further enhance the performance, a novel scheme based on contour reconstruction is also provided. This scheme enables the extraction of a closed outermost boundary from a set of fragmented object points, and represents this boundary as a one-dimensional sequence. The similarity between a pair of fragmented boundaries can then be determined by searching three corresponding point pairs in the one-dimensional sequences. This reduces the dimensions of the problem to three (instead of six for the original set of affine parameters). Experimental results show that the proposed method is considerably faster than previous schemes, and can realise a high success rate in identifying matched contours. Furthermore, a method known as labelled chamfer distance transform (LCDT) is proposed to improve computational efficiency in contour reconstruction. LCDT enables faster generation of the distance image and correspondence map. Compared with its parent scheme, the proposed LCDT approach achieves up to an order of magnitude of acceleration in the entire matching process. Apart from matching in a noise-free background, an attempt to match shapes under a noisy setting is made, in which a scheme called successive erosion and distance accumulation (SEDA) is proposed to alleviate the sensitivity of the distance images to noise. Experimental evaluation demonstrates the SEDA scheme can achieve a high success rate in identifying matched contours under moderate amount of noise contamination.
Online Catalog Link:
Appears in Collections:EE - Doctor of Philosophy

Files in This Item:

File Description SizeFormat
abstract.html132 BHTMLView/Open
fulltext.html132 BHTMLView/Open

Items in CityU IR are protected by copyright, with all rights reserved, unless otherwise indicated.


Valid XHTML 1.0!
DSpace Software © 2013 CityU Library - Send feedback to Library Systems
Privacy Policy · Copyright · Disclaimer