一种新的用于三维检索的快速邻域搜索方法  

USING FAST NEIGHBOR QUERY IN 3D MODEL RETRIEVAL

作　　者：李晓方[1] 武仲科[1] 樊亚春[1] 周明全[1] 柳勇光[1] 

机构地区：[1]北京师范大学信息科学与技术学院,北京100875

出　　处：《北京师范大学学报（自然科学版）》2011年第5期473-477,共5页Journal of Beijing Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(61001168/F010403);国家"863"计划资助项目(2008AA01Z301)

摘　　要：在传统的层次聚类算法上,对孤立聚类进行特殊处理,并在此基础上提出了与传统内容检索方法不同的三维模型索引检索技术.首先提取三维模型的深度图像特征并将特征值存入数据库;其次利用改进的层次聚类算法对特征值进行聚类并计算聚类中心值;之后计算每个三维模型内容特征与其所在聚类中心的距离并保存到数据库中,查询时计算待查询模型特征与各个聚类中心的距离并排序;最后在每个聚类里面进行检索,与聚类中心的距离位于待检索模型与该聚类中心距离的邻域就是查询结果.在该方法中,将每个模型特征与其所在聚类中心的距离作为其索引.实验结果表明,该方法在保证检索质量的基础上大大提高了检索速度.With continuous improvement of high-dimensional indexing technology and massive increase in three-dimensional models,content-based three-dimensional model retrieval is facing various challenges.Therefore a different three-dimensional model retrieval index was proposed,based on traditional hierarchical clustering algorithms and special dealing with isolated clusters.The features of each model,distance from center of cluster with which the model was specified,were all stored in a database.The database was then searched.Distances from characteristics of query model to centers of all clusters were calculated,distances were then sorted in an ascending order.Loop search was then done in each cluster;when query models inside cluster were located in the neighborhood of the distance between query model and center of cluster,the returning results of retrieval were determined to be successful.In this method,distance of model feature to center of cluster was stored in the database as an index for that model.Experimental data showed that this method improved greatly retrieval speed without compromising search quality.

关 键 词：三维检索 聚类 高维 邻域 

分 类 号：TP393.092[自动化与计算机技术—计算机应用技术]