Clustering based on scale-space theory /

Clustering is defined as the partitioning of unlabeled data

Bibliographic Details
Main Author: Nakamura, Eiji
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1997.
Subjects:
Online Access:http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=739891291&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:Clustering is defined as the partitioning of unlabeled data
samples into subgroups
with similar characteristics. The study of clustering or
cluster analysis has been an interdisciplinary research
effort and hence numerous clustering algorithms have been
reported in the literature. In clustering algorithms, the
number of clusters or groupings is usually assumed to be
known or given. This dissertation presents a novel
clustering algorithm, called muti-scale clustering, by not
making such an assumption regarding the number of clusters.
The developed algorithm utilizes scalespace theory to form
clusters. The use of this theory allows an optimum number
of clusters to be found based on survival or existence of
clusters over many scale levels. In other words, an optimal
number of clusters as well as optimal locations of cluster
prototypes and membership values are found in an objective
manner by defining lifetime and drift speed clustering
criteria. Due to the non-parametric nature of this
algorithm, different cluster types can be handled without
specifying a norm inducing matrix. The scope of this
algorithm is addressed by enhancing the following two
cluster-embedded problems (i) the expectation-maximization
technique for mixture density estimation in pattern
recognition, and (ii) the Hough transform technique for edge
detection in image processing.
Item Description:Vita.
"Major Subject: Electrical Engineering".
Physical Description:xi, 113 leaves : illustrations ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references: pages 106-112.