Image Segmentation: A Novel Cluster Ensemble Algorithm
Cluster ensemble has testified to be a good choice for addressing cluster analysis issues, which is composed of two processes: creating a group of clustering results from a same data set and then combining these results into a final clustering results. Ho
- PDF / 873,632 Bytes
- 8 Pages / 439.37 x 666.14 pts Page_size
- 33 Downloads / 246 Views
)
College of Computer Science and Technology, Harbin Engineering University, Harbin 150001, China [email protected]
Abstract. Cluster ensemble has testified to be a good choice for addressing cluster analysis issues, which is composed of two processes: creating a group of clustering results from a same data set and then combining these results into a final clustering results. How to integrate these results to produce a final one is a significant issue for cluster ensemble. This combination process aims to improve the quality of individual data clustering results. A novel image segmentation algorithm using the Binary k-means and the Adaptive Affinity Propagation clus‐ tering (CEBAAP) is designed in this paper. It uses a Binary k-means method to generate a set of clustering results and develops an Adaptive Affinity Propagation clustering to combine these results. The experiments results show that CEBAAP has good image partition effect. Keywords: Cluster ensemble · Binary k-means · Adaptive affinity propagation clustering · Image segmentation
1
Introduction
Image segmentation can be defined as a process that partitions an image into several non-intersected areas and each area is homogeneous and related. The relevant section is the essential topic in computer vision. [1, 2]. Since color image contains more image information than gray one. Academics have been paying more attention to color image segmentation. They have developed many image segmentation algo‐ rithms based on clustering [3]. In essence, clustering analysis belongs to the problem of an unsupervised pattern recognition. It can be viewed as a process of clustering and unmarked data points are devided into k groups with a few clustering criteria so that the intercluster dissimilarity is maximized while the intracluster dissimilarity is minimized [4]. The goal of cluster analysis is to capture the underlying structure of specific properties in the data and several clustering criteria. Many kinds of clus‐ tering approaches have been presented over the past few decades, such as Expecta‐ tion Maximization (EM), k-means, graph-based approaches and hierarchical clus‐ tering approaches like Single-Link, Fuzzy c-Means [5, 6]. However, a clustering approach that can accurately capture the underlying struc‐ ture of all data sets has not been proposed yet. It imposes an organization to the data following the data, the characteristics of an utilized dissimilarity function and an internal standard after we apply the clustering approach to a data set. The idea of
© Springer Science+Business Media Singapore 2016 W. Che et al. (Eds.): ICYCSEE 2016, Part I, CCIS 623, pp. 410–417, 2016. DOI: 10.1007/978-981-10-2053-7_36
Image Segmentation: A Novel Cluster Ensemble Algorithm
411
integrating different clustering results becomes an alternative pattern. The clustering methods of improving the quality are the results which are challenging. The successful combination of supervised classifiers is the foundation of cluster ensemble. When it is given a group of objects, a cluster ensemble
Data Loading...
