面向可变权值的多特征索引结构  

Multi-feature Index Structure for Weighted Query Applications

作　　者：何云峰[1] 于俊清[1] 

机构地区：[1]华中科技大学计算机科学与技术学院,武汉市珞喻路1037号430074

出　　处：《武汉大学学报（信息科学版）》2010年第8期920-924,共5页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金资助项目(60703049);武汉市青年科技晨光计划资助项目(200850731353);华中科技大学自主创新基金资助项目(M2009019)

摘　　要：在基于样例的视频检索中,视频数据采用多个高维特征数据描述,针对不同的检索应用中这些特征数据的权值经常会发生变化的情况,提出了一种面向可变权值的多特征索引树(multi-feature index tree)结构,以满足用户在样例检索过程中对特征权值进行自定义的设置。多特征索引树采用适应于浏览的树型结构对视频的多个特征向量进行索引,检索时,通过遍历最低一层的集合节点,以减少数据维数对检索效率的影响,并针对多特征索引树结构,提出了一种快速确定检索距离值的ADD-kNN检索算法。实验表明,这种索引结构及相应的检索算法具有较好的性能。In the application of video retrieval by sample,video data is described by multiple high-dimensional features,and the weights of these features are changed in different queries.We propose a new indexing structure called multi-feature index tree（MFI-Tree） to index multiple high-dimensional features of video data for this retrieval application.MFI-Tree employs tree structure which is benefit for browsing application,and travels the last level aggregate node in retrieval application to improve the performance.And more,aggressive decided distance for kNN search algorithm which fast reduces the distance to prune the search space more effectively is proposed.The experimental results show that MFI-Tree and ADD-kNN algorithm offer performance advantages over sequential scan.

关 键 词：多特征索引树 视频索引 权值 k-NN检索 

分 类 号：P208[天文地球—地图制图学与地理信息工程]