基于度量空间的动态高维索引结构分析

本文对模式识别中的经典FCM聚类进行了改进,对经典高维索引结构进行了分析,并将这种改进的FCM算法同树形索引结构相结合,提出了一种新的基于度量空间的动态高维索引结构HC-Tree(Hierarchical clustering tree)。插入算法保持HC-Tree更新时的动态平衡、分裂条件和分裂算法,使HC-Tree具有更紧致均匀的节点,大大减少了重叠,并实现了K近邻查询和范围查询。利用图像数据库特征向量进行了测试,结果表明,HC-Tree性能优于M-Tree和Slim-Tree。

Analysis on dynamic high-dimensional indexing structure based on metric space

This paper modifies the traditional fuzzy c-means(FCM),and gives a analysis of high-dimensional indexing structure.Combining the modified FCM to tree-like indexing structure,this paper introduce a novel dynamic high-dimensional indexing structure based on metric space,called HC-Tree.The insertion methods of HC-Tree make the tree balance and dynamic.We give a method to determine whether a node reach the qualification of splitting operation.Then we give the methods of nodes splitting strategy,which make the nodes of HC-Tree are much more compact,uniform and symmetrical,and then make much less overlaps.We implement the K-NN queries and Range queries.The experimental results show that the HC-Tree outperforms the M-Tree and Slim-Tree with more efficient retrieval and browsing.