||The stochastic watershed method, proposed by Angulo and Jeulin in 2007, is a powerful tool for measuring the strength of boundaries in an image. The method computes, for each piece of boundary in the image, the probability to appear as a contour in seeded watershed segmentation with randomly selected seeds. Contours that appear with high probability are considered to be more important.
In the original publication by Angulo and Jeulin, the stochastic watershed probabilities were estimated by Monte Carlo simulation, i.e., repeatedly selecting random seedpoints and performing watershed segmentation. In this talk, I will present an efficient method for computing the probabilities exactly, with out performing any Monte Carlo simulations. In practice, the proposed method is faster than any previously reported method by more than an order of magnitude.