Determining the Convex Hull in Large Multidimensional Databases

Determiningthe convex hull ofa point set isa basic operation for many applications of pattern recognition, image processing, statistics, and data mining. Although the corresponding point sets are often large, the convex hull operation has not been considered much in a database context, and state-of-the-art algorithms do not scale well to non main-memory resident data sets. In this paper, we propose two convex hull algorithms which are based on multidimensional index structures such as R-trees. One of them traverses the index depth-first. The other algorithm assigns a priority to each active node (nodes which are not yet accessed but known to the system), which corresponds to the maximum distance of the node region to the tentative convex hull. We show both theoretically as well as experimentallythatouralgorithmsoutperformcompetitivetechniquesthat do not exploit indexes.

• Type: Conference
• Year: 2001
• Where: DAWAK
• Authors: Christian Böhm, Hans-Peter Kriegel
• Keywords: Convex Hull, Convex Hull Ofa, Convex Hull Operation, DAWAK 2001, Information Management