Component labelling on a MIMD multiprocessor

Abstract We present a parallel component labelling algorithm that is based on a "divide-and-conquer" strategy and is suited for coarse-grained MIMD parallel machines. The complexity of the algorithm is analyzed and timings on an Intel iPSC/2® hypercube are given. Furthermore, we measure the load imbalance for a number of images and establish general theoretical upper bounds for it. We also indicate how a good static load balance can be achieved.