Skeletal-based microstructure representation and convolution reconstruction

This paper introduces the skeletal-based microstructure representation. The skeletal representation allows for compact storage while capturing the main characteristics of the microstructure; and the separation of geometry and topology. The skeletal representation consists of two elements: primitive-based skeleton, representing topology, and a kernel representing the shape of the primitives in the skeleton. We supplement the representation with three operations: skeletonization, primitive identification, and convolution of the skeleton with a kernel. To demonstrate the robustness of the representation, we perform a quantitative study to show the decrease in the number of parameters needed to represent the several types of microstructure with respect to the voxel-based representation. We report a low error of reconstruction of the proposed representation method.