Medial axis extraction and shape manipulation of solid objects using parabolic PDEs

Abstract

Shape skeletonization (i.e., medial axis extraction) is powerful in many visual computing applications, such as pattern recognition, object segmentation, registration, and animation. This is because medial axis (or skeleton) provides more compact representations for solid models while preserving their topological properties and other features. Meanwhile, PDE techniques are widely utilized in computer graphics fields to model solid objects and natural phenomena, formulate physical laws to govern the behavior of objects in real world, and provide means to measure the feature of movements, such as velocity, acceleration, change of energy, etc. Certain PDEs such as diffusion equations and Hamilton-Jacobi equation have been used to detect medial axes of 2D images and volumetric data with ease. However, using such equations to extract medial axes or skeletons for solid objects bounded by arbitrary polygonal meshes directly is yet to be fully explored. In this paper, we expand the use of diffusion equations to approximate medial axes of arbitrary 3D solids represented by polygonal meshes based on their differential properties. It offers an alternative but natural way for medial axis extraction for commonly used 3D polygonal models. By solving the PDE along time axis, our system can not only quickly extract diffusion-based medial axes of input meshes, but also allow users to visualize the extraction process at each time step. In addition, our model provides users a set of manipulation toolkits to sculpt extracted medial axes, then use diffusion-based techniques to recover corresponding deformed shapes according to the original input datasets. This skeleton-based shape manipulation offers a fast and easy way for animation and deformation of complicated solid objects.