Topology-preserving Network Snakes

Active contour models have been part of extensive research since their introduction in the late eighties, because their concept of coupling the image data with shape control is a powerful method to delineate non-rigid curves. Various applications and enhancements have been examined using this technique, for example the development of multiple and coupled active contour models. Latest research deals with the introduction of topology into the model of active contours: network snakes exploit the given topology during the energy minimization process leading to improved results concerning the delineation of arbitrary networks and object borders of adjacent objects. However, the utilization of topology requires obviously a given correct topology and, moreover, presumes the preservation of this initial topology over the period of optimizing the energy functional. This fact can not be guaranteed in general, because close contour parts can merge or nodes with higher degrees can move around each other. Thus, a new topologypreserving energy is introduced in this paper to enhance the procedure of network snakes. The preservation of the topology is demonstrated with a synthetic example and with real application scenarios, in particular, the refinement of road networks in aerial images. The results demonstrate the general functionality and transferability of the proposed new topology-preserving energy. Concluding remarks are given at the end to point out further investigations.