Fast Object Segmentation by Growing Minimal Paths from a Single Point on 2D or 3D Images

摘要: In this paper, we present a new method for segmenting closed contours and surfaces. Our work builds on variant of the minimal path approach. First, an initial point desired contour is chosen by user. Next, keypoints are detected automatically using front propagation We assume that object has boundary. This a-priori knowledge topology used to devise relevant criterion stopping keypoint detection propagation. The final domain visited will yield band surrounding interest. Linking pairs neighboring with paths allows us extract from 2D image. approach can also be finding open curve giving extra information as criteria. Detection variety objects real images demonstrated. Using similar idea, networks 3D image called Geodesic Meshing. proposed applied data promising results.