VA_Trie_一种用于近似k近邻查询的高维索引结构.pdfVIP

计算机研究与发展 ISSN CN 111777/ TP Journal of Computer Research and Develop ment 42 ( 12) : 22 13～22 18 , 2005 VATrie : 一种用于近似 k 近邻查询的高维索引结构 董道国　刘振中　薛向阳 (复旦大学计算机科学与工程系 　上海 　200433) (roddong @gmailcom) VATrie : A Ne w and Eff icient High Dimensional Index Structure f or Approxi mate k Nearest Neighbor Query Dong Daoguo , Liu Zhenzhong , and Xue Xiangyang ( ) Dep art ment of Comp uter S cience an d Eng ineering , Fudan University , S hang hai 200433 Abstract 　Since 1990 ’s , great p rogress has been made in t he area of contentbased multimedia information ret rieval A very challenging p roblem emerged at t he same time : how to organize high dimensional vector s so t hat efficient similarity query could be realized Many index st ruct ures have been p ropo sed to solve t his p roblem , such as RTree and it s variant s , VAFile , ATree etc From t he p ublished result s , it can be con cluded t hat mo st of t hese met hods could achieve good query performance when t he dimensionalit y is less t han 20 However , t he performance suffer s greatly as t he dimensionalit y increases To obt ain efficient sim ilarit y query in higher dimensional sp aces , a new index st ruct ure called VATrie is int roduced The key idea behind VATrie is adopting t he idea of quantization to comp ress t he vector s and t hen employing t he Trie st ruct ure to organize and manage t he app roximations The experiment al result s show t hat VATrie outper forms ATree and sequential scan in high dimensional sp aces Key words 　index st ruct ure ; similarity query ; information ret rieval 摘 　要 　近年来 ,随着多媒体信息检索技术的不断发展 ,如何实现高维特征矢量的快速相似性查询成为 一个重要的研究课题 为此 ,人们提