PD-Tree:Novel High-dimensional Indexing Structure on a Mapping Space
Liang Yu · Journal of Chinese Computer Systems · 2011
It is acted as the most effective method for lowing computation times of distance function to filter data space efficiently.A novel filtering technique for multidimensional data is proposed here-on the basis of space mapping,the dimension with the biggest combined-variance is selected as the principal dimension(PD),and the unrelated data is filtered with it combining with the triangle inequality.Then a new multi-level index structure based on the principal dimension filtering technique,called PD-Tree,is presented.In order to show the property-kept degree of the principal dimension,variance-coverage weight is calculated and analyzed.Finally,all experiments in varied data scale show that PD-Tree is superior in lowering computational times of distance function,reducing the cost of CPU and improving retrieval speed.