Polygonal approximation of image contours on parallel machines

Abstract

In this paper, we present the parallel implementation of an algorithm for polygonal approximation of image contours-one of the steps involved in image processing for two-dimensional machine vision. The proposed algorithm proved to be a very good compromise between performance-typical of fast, scan-along (i.e., single-pass) algorithms-and quality-usually maximized by slow, iterative algorithms. A challenging task has been the implementation of this algorithm on a distributed-memory MIMD machine, constituted by a transputer network, since the necessity of considering data from a "global" point of view tends to overcharge the inter-processor communication structure. The proposed implementation of such an algorithm led to a satisfactory behaviour on small number of processing nodes, but showed also the need of a sophisticated protocol for information interchange among processing nodes, in order to overcome scalability limits.<>