Author(s): Zhenping Xie | Shitong Wang | Dian You Zhang | F.L. Chung | Hanbin
Journal: Information Technology Journal
ISSN 1812-5638
Volume: 6;
Issue: 4;
Start page: 541;
Date: 2007;
VIEW PDF
DOWNLOAD PDF 
Original page
Keywords: Enhanced possibilistic clustering method | outliers | flexible hyperspheric partition | image segmentation
ABSTRACT
The possibility based clustering method PCM (possibilistic clustering method) was first proposed by Krishnapuram and Keller to overcome FCM for noises and outliers. However, it still has the following weaknesses: 1) the clustering results are dependent on parameter selection and initialization; 2) the outliers cannot be labeled in a reasonable way. In this study, in order to avoid the above weaknesses, a novel modified PCM version, called EPCM (Enhanced PCM), is presented. First, a novel strategy of Flexible Hyperspheric Partition (FHP) is proposed and then, this strategy is used to construct the objective function of EPCM with some novel constraints. The main advantage of EPCM is that it can label the outliers adaptively and accurately, which enhances the clustering performance and increases its potential applications. Our experimental results about artificial datasets and image segmentation confirm the above standpoints.
Journal: Information Technology Journal
ISSN 1812-5638
Volume: 6;
Issue: 4;
Start page: 541;
Date: 2007;
VIEW PDF
DOWNLOAD PDF Original pageKeywords: Enhanced possibilistic clustering method | outliers | flexible hyperspheric partition | image segmentation
ABSTRACT
The possibility based clustering method PCM (possibilistic clustering method) was first proposed by Krishnapuram and Keller to overcome FCM for noises and outliers. However, it still has the following weaknesses: 1) the clustering results are dependent on parameter selection and initialization; 2) the outliers cannot be labeled in a reasonable way. In this study, in order to avoid the above weaknesses, a novel modified PCM version, called EPCM (Enhanced PCM), is presented. First, a novel strategy of Flexible Hyperspheric Partition (FHP) is proposed and then, this strategy is used to construct the objective function of EPCM with some novel constraints. The main advantage of EPCM is that it can label the outliers adaptively and accurately, which enhances the clustering performance and increases its potential applications. Our experimental results about artificial datasets and image segmentation confirm the above standpoints.
