Return to search

Fast 2D shape approximation algorithms and their errors

In this thesis, efficient algorithms for fast shape approximation in the area of machine and robot vision are analysed. The core of the thesis consits of two main parts, an error analysis of chain coded silhouettes and a simplified, hence fast shape approximation scheme for practical applications. The area of shape description and approximation is introduced. General techniques for processing of boundary data are outlined followed by a brief description of the image processing system which was developed during the study. Processing procedures developed and their options are presented. Two applications realized by the author have emerged from the algorithm development, firstly a semi-automatic shape registration and measurement procedure and secondly, a low-cost robot vision system. The advantage of the experimental robot vision system is an effective communication between the robot and the object recognition system. The shortcomings of these applications, mainly inaccuracy of the shape description lead to the introduction of an error analysis for boundary descriptors. In contrast to existing methods, the error analysis can be applied to straight lines, circular arcs, and arbitrary shapes consisting of these two shape primitives. The analysis is comprehensive and covers all kinds of chain code sets representing various pixel shapes as well as different pixel configurations. The results of the analysis lead to the development of simple ways of overcoming the accuracy limitations of conventional methods. Two algorithms for improving the length estimation of object outlines are introduced. A comparison with existing methods showing the effectiveness of these algorithms is made. For practical applications of object recognition, a new, simple and hence fast and effective algorithm for polygon approximation, called the "arc operator" is developed. The combination of the arc operator and the length correcting algorithm directly improves the ability to identify partially visible shapes. The performance of the arc operator is compared to six other well known algorithms. Slight disadvantages in accuracy are outweighed by the enormous advantage in processing time, particularly with an integer implementation. An example of overlapping parts demonstrates the potential of the arc operator to extract characteristic shape descriptions.

Identiferoai:union.ndltd.org:ADTP/275034
Date January 1989
CreatorsFriedrich, W. E.
PublisherResearchSpace@Auckland
Source SetsAustraliasian Digital Theses Program
LanguageEnglish
Detected LanguageEnglish
RightsItems in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated., http://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm, Copyright: The author

Page generated in 0.0011 seconds