SGER: Data-Distribution Properties in High Dimensional Euclidean Space and their Applications in Optimizing Multi-Media Database Accesses

点击翻译按钮获取中文摘要

The characteristics of data-distribution in high dimensional Euclidean space play a fundamental role in many areas of computer science, including database systems, pattern recognition and multimedia. Several basic data distribution properties, namely the angle, the distance and the radius properties in high dimensional Euclidean Space are being investigated. Both real data sets such as feature vectors for image data and synthetic data sets with distributions such as normal, zipf, self similar and fractal are being used to analyze these properties. A component of this research is to exploit these basic properties in developing efficient indexing schemes for nearest neighbor search. Significant performance improvement in search time in the proposed approach over those currently available is expected in this research. In addition, there is a plan to apply the indexing scheme to commercial image database systems.