Article Preview
TopTaking different approaches, the watershed transform is defined diversely on the literature, each of which producing a different set of solutions (Vincent & Soille, 1991; Meyer, 1994; Lotufo & Falcão, 2000; Bieniek & Moga, 1998; Cousty et al., 2009). The definitions are based on regional or global elements, such as influence zones and shortest-path forests with max or sum of weights of edges, or on local elements, such as the steepest descent paths, where the neighbour’s information is used to create a path to the corresponding minimum. These definitions are thoroughly revised on the literature (Audigier & Lotufo, 2007; Roerdink & Meijster, 2000; Cousty et al., 2009). In this work the local condition definition is used, henceforth named LC-WT (Local Condition Watershed Transform).
The LC-WT definition was introduced by Bieniek et al. (1997) with the purpose of achieving speedups on the parallel watershed transform, once it mimics the behaviour of a drop of water on a surface, requiring less steps of global processing, and thus requiring less communication and synchronisation. Next, we discuss the algorithms that implement this definition on their sequential and parallel versions.