ND-Tree: Multidimensional Indexing Structure

The importance of multimedia databases has been growing over the last years in the most diverse areas of application, such as: Medicine, Geography, etc. With the growth of importance and of use, including the explosive increase of multimedia data on the Internet, comes the larger dimensions of these databases. This evolution creates the need for more efficient indexing structures in a way that databases can be useful, returning accurate results in a short time. Typically, these databases use multi-dimensional indexing structures to deal with feature vectors extracted from multimedia elements. However, the majority of existing multidimensional indexing structures, suffer from the well-known “dimensionality curse”, making the search in high-dimensional spaces a hard problem. In this work we developed an efficient indexing structure to support large databases containing data of high dimensions (over 100). The new indexing structure, *D-Tree (*orm Diagonal Tree), is based on a new dimension reduction technique based on two metric measures, Euclidean norm and distance to the unity cube diagonal. Experimental results show that our solution is more efficient and that the two metric measures provide a more effective pruning mechanism, while compared to other indexing structures.