Revisiting adaptive patches for mesh denoising

This paper revisits the mesh denoising problem by presenting an extended and refined version of our previous adaptive patch framework. Triangular meshes acquired from real-world data often contain noise that degrades surface quality, and denoising techniques must remove such distortions while preserving true geometric details, a difficult task due to the similar high-frequency characteristics of noise and sharp features. Our revisited approach follows a two-step strategy that first applies bilateral filtering to the normal field using adaptive patches and then updates vertex positions to align the mesh geometry with the filtered normals. The core contribution is the computation of adaptive patches through local quadratic optimization problems that can be controlled to capture local surface structure in both smooth and feature-rich regions. We provide additional analysis, refinements, and experiments that were not included in the original version. Comprehensive evaluations on synthetic and real datasets demonstrate that this revisited algorithm achieves effective noise removal while preserving important geometric features.