Water Flow Based Vessel Detection in Retinal Images

Previous vessel segmentation methods mainly concentrate on the general structure, and often ignore the accuracy, smoothness and continuity of vessel boundaries. A water flow based method is proposed to solve the problem. It embodies the fluidity of water and hence can handle the complex topological changes of vessels. A snake-like force functional combining edge-based and region-based forces produces capability for both accuracy and range. Properties analogous to surface tension and adhesion are also applied so that the smoothness of the evolving contour and the ability to flow into narrow branches can be controlled. The technique has been assessed on synthetic and real images, and shows excellent detection performance and ability to handle noise.