DMS 15-721 Project Report

Indexing structures of modern multimedia databases almost always use a predefined measure of distance, which is usually independent of the data. However, knowing how a given data point relates to other data points in the database can often yield insight into what data points should be considered similar. Computationally expensive data mining operations allow users to classify data into clusters. However, these operations require several passes through data, and classifications can be made obsolete when data is inserted, deleted, or modified. As a response to this problem, we developed and compared two indexing schemes that facilitate cluster-based queries.