Geometric Active Contours Implementation On Point Clouds

Image segmentation is a process of separating the target contour after marking the objects of the background according to the prior knowledge. Until now, there is no general segmentation algorithms, and no criteria can be consulted to choose appropriate segmentation methods, so the challenging problem of effectively separating the target we are interested in from the complex background draws more and more attention.Geometric active contours based on level set have made great progress on the area of image segmentation in recent years, since it solved some problems parametric deformable models failed, such as automatically changing the topology, dealing with large deformation. Firstly, a new geodesic active contour based on adaptive neighboring influence is introduced. It is capable of robustly dealing with the leakage issue by enlarging the capture area through reconstructing the internal energy and using GVF as the external energy. Then we present another geometric active contours implemented on point clouds based on Mumford-Shah. In this method, a set of points that are adoptively distributed according to the image feature are created, and then the meshfree is used to solve the problem for the purpose of getting higher accuracy. The initial curve could be anywhere in the image and the interior contours are automatically detected. At last we present the texture segmentation based on Mumford-Shah with Gabor filter.