Extract feature curves on noisy triangular meshes

Abstract Feature curve is a powerful shape descriptor for a wide spectrum of applications, such as shape registration, geometry compression and surface reconstruction in CAD and reverse engineering. However, feature curve extraction from noisy data has been a challenging problem for decades. In this paper, we propose a robust algorithm for extracting feature curves from noisy triangular meshes. Specifically, we first slice the triangular mesh into a series of sections along a given slicing guide line interactively stroked by users. Under different levels of noise frequencies and amplitudes, we measure the similarities of adjacent sections and extract their corresponding salient points, followed by connecting the consistent salient points to form feature curves. Instead of exploiting the local geometric information as previous methods do, we leverage the global geometric properties of section curve pairs of the input mesh to extract feature curves. This enables our method to cope with a high level of noise. Moreover, our method automatically computes optimal noise frequencies and amplitudes, which makes feature curve extraction less sensitive to different levels of noise than others. A variety of experiments demonstrate the robustness and effectiveness of our proposed method on low-quality mesh data.