Visibility-driven skeleton extraction from unstructured points

摘要

We present an algorithm for curve skeleton extraction from unstructured points using visibility as a guide. We introduce visible cells inside the point samples. By locating a viewpoint in the space, visible points from the viewpoint are connected to indicate the visible region, which locally captures the interior structure of the points and is called a visible cell. We then analyze and clean the visible cells for noisy and incomplete points. By connecting the cleaned visible cells with a region growing method, we obtain the skeleton along which the visibility of the unstructured points is balanced. No normal computation or ellipse fitting is required by using the proposed method. We show that the proposed skeleton well represents the shape. We experiment with our algorithm on imperfect points and show that comparative curved skeletons are extracted with relative lower computational cost.