Active contours driven by divergence of gradient vector flow

Gradient vector flow (GVF) is an important and widely used external force field for snake evolution. Due to difficulties in evolving over saddle points and stationary points of GVF field, snakes based on GVF suffer from a poor performance of dealing with complex geometries. In this paper, we investigate and analyze the characteristic of the divergence of GVF field and the flux of GVF field across the curve, to the end propose a geometric active contours model. In the new model, the external driving force just is the negative gradient of an energy functional. The proposed model greatly improves the active contours in dealing with complex geometries, and has good robustness and a wide capture range. In addition, from the differential geometry point of view, we give a uniform selection of the edge map used in computing GVF field for either a scalar or a vector-valued image. It makes the proposed model be implemented straightforwardly for both kinds of image. Various experiments are provided to demonstrate the capability of the proposed model in image segmentation and object recovery, especially when complex geometries are dealt with. HighlightsA new geometric active contours model is proposed.The divergence of gradient vector flow (GVF) is investigated and used as a criterion in image segmentation.The proposed model largely solves the problem of GVF-based snakes associated with dealing with complex geometries.A uniform selection of the edge map used in computing GVF field is presented.