【作者】 唐洁；

【导师】 王天江；

【作者基本信息】 华中科技大学 ， 计算机应用技术， 2008， 硕士

【摘要】 随着大规模数字图像库的出现,传统的依赖于人工标注进行的基于文本的图像检索技术已经无法满足用户日益增长的要求,基于内容的图像检索技术(Content-based image retrieval CBIR)便应运而生。CBIR的一般做法是提取图像的某些特征,构成其特征向量,为方便检索,对特征空间建立索引。现在已有一些不同的建立索引的方法,CM-tree(度量聚类树)是一种较新的方法。CM-tree节点中保存了聚类半径及聚类之间距离表,基于度量空间聚类的范围查询算法利用距离表的信息,根据三角不等式减少距离计算的次数,加快检索的时间。在处理单个的、小数据量的查询检索上,基于度量聚类检索具有一定的优势。但是用户在一段时间内提交的查询是具有相当的关联性的,或者极端的来说是重复性的,如果仍然使用常规的检索方式对度量聚类索引树进行检索,毫无疑问的会多做许多重复的工作,如对同一张图片进行多次检索,或者对关联性相关很大的图片也是进行多次的检索。基于联盟对度量聚类检索的优化方法可以解决上述问题。联盟即主查询和被邀请查询经过一系列的规则所形成的复合查询的技术手段。联盟的应用层次就在度量聚类树根节点下的第一层导航节点上,在此层次上,主查询在不同的导航节点上分别和被邀请查询根据一定的规则进行联盟操作,并创建复合查询,对复合查询进行相应节点上的检索。同时,在主查询和被邀请查询相应的属性中保留在该节点上的联盟信息,被邀请节点在下一次查询的过程中就不需要对有联盟信息的节点上进行检索,此操作可避免重复查询,减少查询的次数,提高查询的效率。基于联盟对度量聚类检索优化,主要在批量数据、实时查询的背景下应用,实验表明了采用联盟的技术能有效地提高查询的效率,减少重复查询的次数。更多还原

【Abstract】 With the emergence of the large-scale digital image database, the traditional reliance on artificial mark the text-based image retrieval technology has been unable to meet the growing requirements of users, so content-based image retrieval technology (Content-based image retrieval CBIR) appears. Generally, CBIR extracts certain features of images; its characteristics pose a vector, to facilitate the retrieval of the feature space index. Now there are different ways of indexing, CM-tree (clustered metric tree) is a relatively new approach.The nodes of CM-tree preserve the clustered radius and the distance tables of clusters, the range query algorithm based on clustered metric speed up the retrieval time by using the distance tables’information, according to the number of calculation reduced by the triangle inequality. The retrieval based on clustered metric have certain advantages when dealing with single query and query of small amount data. However, the queries submitted by users in a certain period of time have considerable relevance, extremely, they’re repetitive. Much have to be done if conventional retrieval method is still adopted, such as retrieval a same picture repeatedly, or retrieval quite related pictures for many times. The optimization method of clustered metric based on coalition can solve the above problem.Coalition is a technical means that a main query and certain invited queries form a new compound query by using some rules. The application layer of coalition is under the root node, the main query invites other queries and conduct coalition operations with certain rules in different navigation nodes of the level, then create a new compound query and retrieval it in the corresponding navigation node. At the same time, the attributes in the main query and the invited queries keep the coalition information, the invited queries need not retrieval in the nodes which keep the coalition information during the next retrieval process, and the operations can avoid retrieving repeatedly, reduce the retrieval times, so it can improve the efficiency of retrieval. The optimization for clustered metric retrieval based on coalition mainly used in the condition of real-time query context. The experiment proved that the usage of technology can greatly improve the efficiency of retrieval.更多还原