Color segmentation based on adjustable fuzzy neural networks and high level understanding of maps

Zhong, Ning (2004) Color segmentation based on adjustable fuzzy neural networks and high level understanding of maps. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (11MB)

Abstract

To digitize and record electronically the huge collection of topographical maps, development of a framework for computer-based color segmentation and high-level map understanding has shown an increasing importance. -- Color map segmentation requires special treatment with respect to line features. However, few research efforts have been put into this area. A novel approach of color segmentation based on adjustable fuzzy neural networks is proposed. The approach is capable of capturing pixel variations along thin line features. Based on a physics model, two heuristics have been derived, which suggest certain pixel variation behaviors in overlapping and boundary areas respectively. Fuzzy neural networks combined with self-adjustment components have been developed. The self-adjustment components dynamically adjust sample pixels among different sample clusters in neural network training. A feature that distinguishes this method from previous supervised neural computing methods is that, by adoption of self-adjustment architectures, it does not strictly require that all samples should have their desired outputs given in advance. Experiments show that the developed method can produce satisfactory segmentation results. -- This study also proposes a novel method for high-level map understanding. A Description Logics (DL) language, GALL(l,D), is introduced to represent spatial objects and their relationships. A set of derivation rules is then proposed to derive a special grammar, called map grammar, from the DL representation. Thus, the map understanding process can be treated as a grammar parsing process. The map grammar is different from traditional grammars in that: (1) it allows for ambiguity; and (2) the input is a collection of primitive map objects instead of an ordered sequence of tokens. A map grammar parsing algorithm based on the Multiple Path Stack (MPS) is proposed. One advantage of this approach is that the knowledge representation is verifiable at design time. In addition, the implementation based on this approach is more robust and highly reusable, since the knowledge representation is separated from the inference mechanism. This research is a first step towards applying DL theory to map understanding. A prototype map understanding system was developed and applied on several test maps. The results show that this method can obtain a satisfactory interpretation of map phenomena.

Item Type: Thesis (Doctoral (PhD))
URI: http://research.library.mun.ca/id/eprint/10035
Item ID: 10035
Additional Information: Bibliography: leaves 251-265
Keywords: Conceptual Modeling, Knowledge Based System, Map Understanding, Description Logics, Syntactic Analysis, Color Segmentation, Fuzzy Logic, Neural Networks, Image Analysis, Pattern Recognition, Feature Extraction, Geographic Information Systems, Meta-Data Modeling
Department(s): Science, Faculty of > Computer Science
Date: 2004
Date Type: Submission
Library of Congress Subject Heading: Geographic information systems; Image analysis; Pattern recognition systems.

Actions (login required)

View Item View Item

Downloads

Downloads per month over the past year

View more statistics