Harmony Search-Based Fuzzy Clustering Algorithms For Image Segmentation
Loading...
Date
2011-02
Authors
Alia, Osama Moh'd Radi
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Fuzzy clustering algorithms, which fall under unsupervised machine learning, are among
the most successful methods for image segmentation. However, two main issues plague these
clustering algorithms: initialization sensitivity of cluster centers and unknown number of actual
clusters in the given dataset. This thesis aims to solve these problems using an efficient
metaheuristic algorithm, known as the Harmony Search (HS) algorithm. First, two alternative
HS-based fuzzy clustering methods are proposed. The aim of these methods is to overcome
the limitation faced by conventional fuzzy clustering algorithms, which are known to provide
sub-optimal clustering depending on the choice of the initial clusters. Second, a new dynamic
HS-based fuzzy clustering algorithm (DCHS) is proposed to automatically estimate the appropriate
number of clusters as well as a good fuzzy partitioning of the given dataset. These
algorithms have been applied to the problem of image segmentation. Various images from
different application domains, including synthetic and real-world images, have been used in
this thesis to show the applicability of the proposed algorithms. Finally, the proposed DCHS
algorithm is applied to two real-world medical image problems, namely, malignant bone tumour
(osteosarcoma) and magnetic resonance imaging brain segmentation. The experimental
results are very promising showing significant improvements compared to other approaches in
the same domain
Description
Keywords
Cluster set theory , Fuzzy algorithms