Study on Contents Based Fast Similarity Search of High-Dimensional Multimedia Data and Its Application

针对高维多媒体数据快速相似检索的索引技术，研究了空间索引结构、降维方法、检索处理算法。提出了一种具有传统表示结构R-树和M-树特征的空间索引结构--MR-树。我们还提出了一种简单映射的枢轴选择方法，这是一种适用于任何度量空间的降维映射。为了处理大量的查询序列，我们可以使用一种新的批处理方法来最小化I/O成本。

As for indexing technique for fast similarity search of high-dimensional multimedia data, spatial index structures, dimension reduction methods, retrieval processing algorithms are investigated. We propose a spatial index structure named MR-tree with both features of traditional representing structures R-tree and M-tree. We also present a pivot selection method for Simple Map, which is a dimension reduction mapping applicable to any metric space. To process a large number of query series, we can use a new batch method to minimize I/O cost.