Parallelization research on watershed algorithm

Watershed algorithm is an effective and efficient image segmentation method. However, due to the influence of noise and fine-grained of flat region, too many local extremum can be detected by algorithm, which easily lead to numerous small regions appearing in the follow-up segmentation and over-segmentation. This paper combines the watershed and seeded region growing algorithm to eliminate the phenomenon of over-segmentation, which gets a better result. But when the image size grows bigger and bigger, the number of region segmented by watershed algorithm and the computation of region merging will both increase sharply, which eventually results in that the process of segmenting image to consume much more time. In order to speed up the segmentation, in this paper a parallel method based on MPI is presented, which is composed of watershed algorithm and region merging. Experiment results show that the given algorithm has a good speedup.