||The stochastic watershed is an increasingly used segmentation algorithm. It estimates a probability density function (PDF) of the most salient contours of the image by Monte Carlo simulations. For this, it repeatedly applies seeded watershed with randomly placed seeds. To receive the final segmentation, the PDF is then processed with a standard watershed approach.
During the last years, we succeeded in improving the performance of the stochastic watershed, both with regard to time and quality of segmentation results: With the exact evaluation of the stochastic watershed (ESW), we provided an efficient algorithm that produces the exact PDF (without Monte Carlo simulations), and with the robust stochastic watershed (RSW), we overcame the problem of false boundaries by adding a small amount of noise to each realization of the stochastic watershed.
Recently, we were working on combining these two methods. If you are interested in how we managed that and want to listen to and discuss some unsolved issues, please attend this Monday seminar.