Fast, parallel watershed algorithm based on path tracing

Abstract

In this paper a fast, parallel watershed algorithm for segmentation of digital grey-scale images is presented. We show an original parallelisation technique based on the "shared nothing" principle and its application to a modified path-tracing watershed algorithm, which allows a vast majority of computations to be broken up into several independent tasks that can be run in parallel on different processing nodes. This approach eliminates the need of any complex synchronization between the threads and translates to a very high efficiency and speed of the algorithm. Sample results are discussed, with emphasis on their correctness and execution times. © 2010 Springer-Verlag Berlin Heidelberg.