Building a Balanced k-d Tree in O(kn log n) Time

TL;DR

This paper presents an efficient method for building balanced k-d trees by presorting data in all dimensions beforehand, avoiding recursive median finding, and enabling parallel construction, with performance benefits in low dimensions.

Contribution

It introduces a presorting algorithm for balanced k-d tree construction that improves efficiency and supports parallel execution, outperforming median-finding methods in low dimensions.

Findings

Comparable performance to median-based methods in four dimensions

Better performance than median-based methods in three or fewer dimensions

Supports parallel tree construction

Abstract

The original description of the k-d tree recognized that rebalancing techniques, such as are used to build an AVL tree or a red-black tree, are not applicable to a k-d tree. Hence, in order to build a balanced k-d tree, it is necessary to find the median of the data for each recursive subdivision of those data. The sort or selection that is used to find the median for each subdivision strongly influences the computational complexity of building a k-d tree. This paper discusses an alternative algorithm that builds a balanced k-d tree by presorting the data in each of k dimensions prior to building the tree. It then preserves the order of these k sorts during tree construction and thereby avoids the requirement for any further sorting. Moreover, this algorithm is amenable to parallel execution via multiple threads. Compared to an algorithm that finds the median for each recursive subdivision, this presorting algorithm has equivalent performance for four dimensions and better performance for three or fewer dimensions.