A Parallel Watershed Algorithm

The watershed transformation is a popular image segmentation algorithm for grey scale images. Sequential watershed algorithms perform a highly data dependent ooding process over the global image. Parallel algorithms which distribute the image over the available processors and simulate the ooding process have a limited, data dependent speedup. This is due to global data dependencies over the sub-domains. For images without plateaus the presented formalism, based on a local condition, leads to a parallel algorithm with almost data independent runtime. Temporary labels are inserted where the solution depends on neighboring results. Then the local watershed segmentation can be solved independently on every sub-domain. The local solutions are merged to a global solution in log P steps on P processors. For images which include plateaus, additional plateau correction steps are necessary. The speedup of the parallel algorithm is veriied on various images.