Hough Transform on the Butterfly and the NCUBE

摘要

This report describes the parallel implementation of the Hough Transform, a technique to detect colinear edge points. Specifically, two contrasting architectures, the Butterfly Parallel Processor, essentially a shared memory machine, and the NCUBE, a direct connection machine in which processors are interconnected in the form of a hypercube are considered. Developing parallel Hough transform algorithms involves addressing questions of optimal processor allocation and parallel peak selection in image neighborhoods. Fast, practical algorithms (subject to inherent lower bounds) are presented, and relevant complexity issues are discussed. Keywords: Parallel implementation; Butterfly parallel processor; Hough Transform.