Fast image segmentation by Watershed Transform on Graphical Hardware

The watershed transform is widely used for image segmentation on computer vision applications. However, sequential watershed algorithms are not suitable for fast applications, once they are one demanding part of several tasks. This paper proposes two parallel algorithms for the watershed transform focused on fast image segmentation using off-the-shelf GPUs. In this sense, these algorithms aims for a speedup by mixing several techniques of the fastest procedures on both sequential and parallel fields. Both algorithms has four major steps, the parallel version processed on SIMD, and the hybrid version mixing parallel and sequential approaches. The experimental results obtained show that the hybrid version is faster, taking advantage of the most appropriate hardware for each task.