TOPOLOGY PRESERVING ACTIVE CONTOURS

Active contours models are variational methods for segmenting complex scenes using edge or regional information. Many of these models employ the level set method to numerically minimize a given energy, which provides a simple representation for the resulting curve evolution problem. During the evolution, the curve can merge or break, thus these methods tend to have steady state solutions which are not homeomorphic to the initial condition. In many applications, the topology of the edge set is known, and thus can be enforced. In this work, we combine a topology preserving variational term with the region based active contour models in order to segment images with known structure. The advantage of this method over current topology preserving methods is its ability to locate boundaries of objects and not only edges. This is particularly useful for highly textured or noisy data.