Computer vision and image understanding vol:63 issue:1 pages:15-26
We present a parallel algorithm for the Euclidean distance transformation (EDT). It is a ''divide-and-conquer'' algorithm based on a fast sequential algorithm for the signed EDT (SEDT). The combining step that follows the local partial calculation of the SEDT can be done efficiently after reformulating the SEDT problem as the partial calculation of a Voronoi diagram. This leads to an algorithm with two local calculation steps with a computational complexity proportional to the number of pixels of the subregions and a global calculation step with complexity proportional to the image perimeter. This article contains a description of the algorithm, a complexity analysis, a discussion on load imbalance, and timings obtained on an iPSC/2. (C) 1996 Academic Press, Inc.