Fuzzy cluster analysis and validation scheme : application to image segmentation퍼지 클러스터 분석 및 검증 기법의 개발 : 영상 분할에의 적용

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 520
  • Download : 0
In this thesis we present a new fuzzy approach to cluster inherently ambiguous data whose boundaries vague and uncertain to classify. To deal with such uncertainties, a spatial homogeneity-based fuzzy c-means (SHFCM) algorithm has been proposed. The SHFCM algorithm is based on two concepts: the initialization using dominant data and the membership updating method using spatial homogeneity. Moreover, a new cluster validity index is incorporated into the SHFCM algorithm to automatically determine the number of clusters for a given data set. The choice of initialization scheme is of importance because the optimal partitions identified by clustering algorithms may vary depending on the initial cluster centroids selected. The proposed initialization method is based on the idea that the dominant data in a given data set are likely to belong to separate clusters. The dominant data are identified using a membership model of data in conjunction with a set of reference data. Data points closest to the dominant data are selected as the initial centroids. Having selecting the initial centroids, the SHFCM algorithm iteratively improves a sequence of sets of fuzzy clusters. The SHFCM algorithm takes the relations between data and clusters into consideration on the spatial space as well as the feature space. The membership degree between a data point and a cluster is influenced by its relations between the spatial neighborhood and the cluster. The influence of neighborhood is discriminated by its homogeneity value. The cluster validity has been used to search for the optimal number of clusters when the number of clusters is not known a priori. A new cluster validity index for the fuzzy c-means-type algorithm has been proposed. The proposed index $υ_os$ introduced two measures in terms of inter-cluster overlap and separation. The inter-cluster overlap measure indicates the degree of overlapping between fuzzy clusters. The separation measure indicates the isolation between fu...
Advisors
Lee, Kwang-Hyungresearcher이광형researcher
Description
한국과학기술원 : 전산학전공,
Publisher
한국과학기술원
Issue Date
2004
Identifier
237659/325007  / 000995034
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 전산학전공, 2004.2, [ vii, 77 p. ]

Keywords

CLUSTER VALIDATION; FUZZY CLUSTERING; Fuzzy CLUSTER ANALYSIS; IMAGE SEGMENTATION; 영상 분할; 클러스터 검증; 퍼지 클러스터링; 퍼지 클러스터 분석; VALIDITY INDEX

URI
http://hdl.handle.net/10203/32856
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=237659&flag=dissertation
Appears in Collection
CS-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0