Electronic Theses and Dissertations





Document Type


Degree Name

Doctor of Philosophy


Computer Science

Committee Chair

King-Ip Lin

Committee Member

Dipankar Dasgupta

Committee Member

Timothy Hnat

Committee Member

Vasile Rus


Searching high dimensional space has been a challenge and an area of intense research for many years. The dimensionality curse has rendered most existing index methods all but useless causing people to research other techniques. In my dissertation I will try to resurrect one of the best known index structures, R-Tree, which most have given up on as a viable method of answering high dimensional queries. I have pointed out the various advantages of R-Tree as a method for answering approximate nearest neighbor queries, and the advantages of locality sensitive hashing and locality sensitive B-Tree, which are the most successful methods today. I started by looking at improving the maintenance of R-Tree by the use of bulk loading and insertion. I proposed and implemented a new method that bulk loads the index which was an improvement of standard method. I then turned my attention to nearest neighbor queries, which is a much more challenging problem especially in high dimensional space. Initially I developed a set of heuristics, easily implemented in R-Tree, which improved the efficiency of high dimensional approximate nearest neighbor queries. To further refine my method I took another approach, by developing a new model, known as R-Forest, which takes advantage of space partitioning while still using R-Tree as its index structure. With this new approach I was able to implement new heuristics and can show that R-Forest, comprised of a set of R-Trees, is a viable solution tohigh dimensional approximate nearest neighbor queries when compared to established methods.


Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.