-
Qijia Huang
Medial Axis as a Bridge Between 3D Representations
19 novembre 2026 - 09:00Salle de conférences IRMA
The medial axis represents a 3D shape through its maximal inscribed spheres and their skeletal structure, combining implicit geometric primitives with explicit connectivity. This hybrid nature makes it a promising bridge between explicit and implicit representations. However, its sensitivity to boundary perturbations makes accurate and compact approximations difficult to construct.
In this talk, we will present two complementary approaches that exploit the medial axis in geometry processing.
First, we introduce a variational formulation that directly fits a set of medial spheres to a target shape. By combining local and global geometric information, our approach achieves a favorable trade-off between approximation accuracy, compactness, and computational efficiency.
We then show how surface reconstruction from unsigned distance fields (UDFs) can be reformulated as a medial axis extraction problem. By extracting the medial axis of an α-offset volume around the zero level set of a UDF, we obtain a unified, grid-free framework for reconstructing accurate and compact meshes of closed and open, manifold and non-manifold, as well as mixed-dimensional geometry. When the input field itself encodes a medial axis, the same approach can also recover that medial axis directly.
Taken together, these methods highlight the potential of the medial axis to bridge explicit and implicit geometric representations through shape approximation and surface reconstruction.
About the speaker : Qijia Huang is a PhD candidate in computer graphics and geometry processing at the ICube laboratory, Université de Strasbourg, supervised by Prof. Dominique Bechmann and Dr. Pierre Kraemer. His research focuses on compact, editable, and structure-aware representations of 3D shapes, particularly medial axis approximation, implicit and hybrid shape representations, and neural distance fields. He develops methods for recovering accurate and interpretable geometric structures, with applications in shape approximation and surface reconstruction. -
Anissa Lamani
TBA
21 janvier 2027 - 09:00Salle de conférences IRMA
TBA