Image Segmentation Based on Viscous Flooding Simulation

The morphological approach to image segmentation is based on the watershed transform. It is a very powerful tool which presents many advantages: it is not parametric, computationally efficient... However, in comparison with energy-based methods as active contours, it does not allow to control the smoothness of the result. In order to introduce geometrical regularization constraints in the morphological segmentation scheme, two options are possible. The first one consists in simulating a viscous flooding for the construction of the watershed. The second one consists in computing the watershed on a smooth relief. In this article, the second alternative is chosen. The relief is modified so that its non viscous flooding is equivalent to the viscous flooding of the original relief. This choice allows to clearly separate the smooth procedure from the strict watershed computation and thereby to preserve the qualities and speed of the watershed transform. Performances of our regularization method are illustrated and discussed on an ultrasound imaging application.