Automatic aerial image segmentation using a modified Chan-Vese algorithm
Xincai Huang, Huang Bai, Sheng Li · 2014
Automatic segmentation of aerial images has been a challenging area of research in recent years. Among numerous image segmentation methods, the level set method has received a great deal of attention which could represent contours or surfaces with complex topology and change their topology in a natural way. The solution of classic level set model, however, can be easily trapped into a local minimum. To overcome this problem, a novel modified dual Chan-Vese model is proposed in this paper. This proposed model is composed of two contours, which evolve towards the edges of objects from inside of the objects and outside of the objects. By reducing the differences between the interior contour and the external contour, the proposed model can partly prevent the solution of the level set method from a local minimum. Experiments show that the proposed model can obtain exact aerial image segmentation.