A massively parallel implementation of the watershed based on cellular automata

The watershed transform is a very powerful segmentation tool which comes directly from the idea of watershed line in geohydrology. It has proved its efficiency in many computer vision application fields. This paper presents a new implementation of the watershed which is optimal according to computation time. The flooding algorithm is reminded. Then, a massively parallel cellular automaton is proposed to propagate data using this approach. We discuss the pros and cons of a hardware implementation and give an example of application. A comparison between the results obtained and theoretical limit cases is also presented.