Online Clustering Using Local Parameter Estimation for Segmentation of Remote Sensing Images

摘要：

This paper proposes an online Clustering paradigm Using Local Parameter Estimation(CULPE) for image segmentation. CULPE adopts the Gaussian kernel function to define the local neighborhood of learning prototypes and produces new prototypes when the membership degrees of new vectors to prototypes are smaller than a preset threshold. We prove that the learning rule of CULPE is of local convergence and CULPE satisfies Bayesian optimal in clustering problems when assuming all prior distributions of unknown clusters are the same. CULPE is unrelated to initial cluster number of data sets and suitable to online analysis of data. Experiments are carried out on two typical data sets: mixed Gaussian distributions and remote sensing images. The results have shown that CULPE has the capacity of online segmenting remote sensing images or clustering data.

关键词： competitive learning(CL) clustering analysis Gaussian kernel function remote sensing image segmentation

作者: Tao Guan Ting Yang Ling-Ling Li

作者单位: Department of Computer Science and Application Zhengzhou Institute of Aeronautic Industry Management Zhengzhou, Henan, China

会议类型: 国际会议

会议名称: 2010 International Conference on Information Security and Artificial Intelligence(2010年信息安全与人工智能国际会议 ISAI 2010)

会议地点: 成都

会议语种:英文

页码: 343-347

在线出版日期: 2010-12-17（万方平台首次上网日期，不代表论文的发表时间）