Stochastic watershed segmentation.
Jesús Angulo, Dominique Jeulin · Biblioteca Digital da Memória Científica do INPE (National Institute for Space Research) · 2007
Abstract This paper introduces a watershed-based stochastic segmentation methodology. The approach is based on using M realizations of N random markers to build a probability density function (pdf) of contours which is then segmented by volumic watershed for defin-ing the R most significant regions. It over-performs the standard watershed algorithms when the aim is to segment complex images into a few regions. Three variants of the random germs framework are discussed, according to the algorithm used to build the pdf: 1) uniform random germs on the same gradient, 2) regionalised ran-dom germs on the same gradient, and 3) uniform random germs on levelled-based gradient. The last algorithm is more complex but it yields the best results.