Skeleton-growing: a vector-field-based 3D curve-skeleton extraction algorithm

Abstract

The vector-field-based method is one of the 3D curve-skeleton extraction algorithms. Typically, critical points in the vector field inside 3D objects are connected to form the curve-skeleton. However, critical points usually do not distribute to all important parts of the 3D object. Therefore, other features are used to produce a reliable result. Although this strategy can deliver a curve-skeleton that captures all of the important parts, the curve-skeleton usually comes with unnecessary segments. This paper proposes the skeleton-growing algorithm that automatically produces the curve-skeleton with small amounts of such segments. It searches for a set of high-curvature boundary voxels as starting points to find a set of suitable seed points that will be used to grow the curve-skeleton. We propose an unnecessary segment removal algorithm that can reduce the skeleton-noise density. A direction-selection algorithm is developed to avoid searching in irrelevant directions. The proposed method can produce a single reliable result curve-skeleton that could be applied in many different applications, including matching, animation, and visualization.