【作者】 徐焕；

【导师】 林坤辉；

【作者基本信息】 厦门大学 ， 计算机应用， 2006， 硕士

【摘要】 随着计算机技术的研究和发展,图像数据库正在许多领域,如医学图像数据库、商标图像库以及数字图书馆等方面得到越来越多的应用。为了在大容量的图像数据库中找到想要的图像,最有效的方法就是实现基于内容检索。这就需要先对图像提取相应特征组成特征库。图像的特征库一般都是多维数据库。多维数据库在计算机图形学,地理信息系统和多媒体数据库等各个领域都有广泛的应用。为了基于内容图像数据库的快速检索,必须借助于高效的索引技术。本文在分析多维数据索引技术的现状和各类索引结构特点的基础上,主要围绕下面两类构造索引的方法进行研究。第一类方法是对多维数据直接构造索引。该类方法是直接对多维空间进行切分来构造树形索引结构。为了尽可能的减少重叠,提高索引效率。对其中的典型索引结构X-Tree、SS-Tree和SR-Tree进行了深入的剖析,针对SR-Tree分裂算法的不足,引入X-Tree中超级节点的思想,对SR-Tree的分裂算法进行了改进,设计了一种新的索引结构ESR-Tree。实验表明,该索引方法能有效提高索引效率。第二类方法是先对多维数据进行处理再构造索引。该类索引算法在构造多维索引之前,先对多维数据通过各种变换,或通过映射降维,或构造矢量压缩文件。为了减少I/O次数和CPU时间,提高检索的效率,对其中的典型算法NB-Tree进行了深入分析,针对其只存储欧氏距离而忽视了空间位置关系的不足,引入角度信息对其进行了改进。实验表明,改进后的算法能有效提高索引效率。更多还原

【Abstract】 Along with the research and development of computer technology, image database has more and more applications in many domains, such as medicine and brand image storehouse, digital library and so on .In order to find the wanted image in the large capacity of image database, the most effective method is content-based retrieval, which needs to extract the corresponding feature of the images and store them as multi-dimensional vector database. multi-dimensional vector database is used in so many domains as the computer graphics, the geographic information system,the multimedia database and so on. For fast retrieval, it must draw support from effective index. Studied on the development process of multi- dimensional index technology and every kind of index structure, the main research and innovation is concerned with the following two methods.The first method is to make index directly on the multi- dimensional vector database. Different space or data demarcation make different index structure. In order to reduce the overlaps and enhance the index efficiency as far as possible, we deeply researched and assay the typical index structure X-Tree, SS-Tree and SR-Tree, and introduce the super node idea in X-Tree to make up for the split algorithm of SR-Tree. A new index structure is designed. Experiments show the method can effectively enhance the index efficiency.The other method is to make some transform, reflection or vector compression to the multi-dimensional data at first, and then make index structure. In order to reduce the I/O number and CPU time, we introduce angle information to NB-Tree to make up for the shortcomings. Experiments show the method can effectively enhance the indexing performance.更多还原