An improved parallel watershed algorithm for distributed memory system

As a classical method of image segmentation in mathematical morphology, the watershed transform has been applied successively into some fields like remote sensing image processing, biomedical and computer vision applications. However the watershed transform is a relatively time consuming task and classical watershed algorithms have a strong recursive nature, so straightforward parallel ones have a very low efficiency. Mekjster and Roerdink (1996; 1995) have proposed an alternative algorithm (M-R for short) which consists of three stages aimed to exploit parallelism fully. The M-R algorithm has much limitation and some underlying logical errors, therefore we present an improved parallel watershed algorithm based on a directed valued graph called components graph for image segmentation. We firstly point out the limits of the M-R algorithm, then describe the theory and steps of our algorithm in detail. Finally, experimental studies and performance measurements are given.