Robustly Extracting Concise 3D Curve Skeletons by Enhancing the Capture of Prominent Features

Out‐of‐core Extraction of Curve Skeletons for Large Volumetric Models

“…This is because our center-oriented face contraction only propagates the fire along the edges of the medial mesh and omits the propagation on the faces to reduce the computational cost. The PF method [CWL22] solves this problem by performing centrality refinement for each skeleton node on the cross-section. However, such a centrality refinement requires global intersection computations, which are computationally expensive for out-of-core models.…”

“…A smaller center deviation value indicates better centrality of the obtained skeleton. The contraction error is introduced in [CWL22] to evaluate the completeness of the skeleton by measuring the terminal contraction of the skeleton. A smaller contraction error means that the skeleton exhibits less contraction at terminations and represents the shape more completely.…”

“…In our experiments, existing methods exhaust the memory and crash when extracting skeletons directly from the three out-of-core meshes. Therefore, for a comprehensive comparison, we select the Buddha mesh which has a complex topology structure to conduct an out-of-core simplification [Lin00], and compare the skeleton extracted by the PF method [CWL22] from the simplified mesh with the skeleton extracted by our method from the high-resolution volume. As shown in Figure 7, the resulted skeletons of the PF method have severely topological errors.…”

“…Similar to the framework in [CWL22] to capture prominent features for high‐quality skeleton extraction with a medial mesh, we also first extract a set of vertices, called prominent vertices , which significantly reduces the data size, and then generate the medial mesh by constructing connections among the vertices to capture the topological structure of the shape, as shown in Figure 4. Instead of extracting prominent vertices from the mesh‐represented medial surface [CWL22], we propose novel measures for extracting prominent vertices from the volumetric distance field, which saves the trouble of extracting the medial surface. We also develop specialized procedures to leverage the volumetric representation for connection construction.…”

“…In this paper, we address the challenges posed by large models based on two observations. First of all, inspired by the work [CWL22] which generates a medial mesh to capture prominent features from the medial surface for high-quality skeleton extraction, we observe that capturing the prominent features considerably reduces the data size and facilitates the process of large models. Besides, since the medial surface is also defined as the singularities of the distance field, we find that the prominent features can be cap-Figure 1: The pipeline of the proposed out-of-core curve skeleton extraction method.…”