Dynamic Clustering-Based Updating Indexing Structure with Triangle Inequality Strategy
Clustering-based indexing structure is one of the most important research directions for high-dimensional data, but they are not dynamic, and the indexing structures have to be reconstructed after adding extra data. To overcome the shortcoming, a two-level indexing method, called PASDS & PPATS method, has been developed by our research. In the PASDS level, clusters and their subspaces can be partially updated, while the indexing trees within the clusters are able to be partially updated at the PPATS level. By choosing proper number of children nodes, the proposed method can balance query accuracy and indexing efficiency. Furthermore, triangle inequality strategy is implemented in query stage by choosing central points (centroids) of clusters as the key points. From experiments, the PASDS & PPATS method has improved considerably the efficiency of updating the whole indexing structures for newly added data, while its k-NN query time and accuracy are similar with existing dynamic indexing methods.
High-dimension data indexing Data clustering updating Triangle inequality strategy CBIR
Ben Wang Hong Huang
College of Computer SoftwareZhejiang University of TechnologyHangzhou, China College of Computer Software Zhejiang University of Technology Hangzhou, China
国际会议
哈尔滨
英文
403-406
2011-01-18（万方平台首次上网日期，不代表论文的发表时间）