用SOM聚类实现多级高维点数据索引

高维点数据的索引是基于内容的信息检索的主要研究问题之一，从SOM聚类算法出发，利用自组织映射的良好性能，解决了R－Tree及其变体算法中的边界索引问题，并能适应维数更高的点数据，同时针对传统聚类算法只能组织一级索引的局限，提出了利用SOM网络组织多级索引，并用半径进行剪枝处理的优化办法，实验结果表明，提出的方法不仅克服了传统聚类方法的搜索过程可能产生的查询错误，而且大大提高了索引的构建和查询效率。

Hierarchical Index of High-Dimensional Point Data Based on Self-Organizing MAP

The content-based multimedia retrieval requires an effective high-dimensional point data index In this paper, a hierarchical index structure is presented, in which the self-organizing map algorithm is employed for data clustering An important proposition of class pruning its corollaries is also proposed And the nearest neighbor and k -NN searching algorithms based on these pruning conditions are also presented The experimental data indicates that the algorithm not only eliminates the possible errors in the query procedure of conventional data clustering methods, but also has very good performance in both index construction and searching