Segmentation-Driven Feature-Preserving Mesh Denoising

TL;DR

This paper introduces a segmentation-driven mesh denoising technique that improves feature preservation by region-wise processing, outperforming existing methods on synthetic and real meshes.

Contribution

The proposed method performs region-wise denoising to better preserve features and can be integrated into existing frameworks, addressing limitations of weight assignment in prior approaches.

Findings

Enhanced feature preservation in denoised meshes

Improved denoising quality on synthetic and real models

Compatible with existing mesh denoising frameworks

Abstract

Feature-preserving mesh denoising has received noticeable attention in visual media, with the aim of recovering high-fidelity, clean mesh shapes from the ones that are contaminated by noise. Existing denoising methods often design smaller weights for anisotropic surfaces and larger weights for isotropic surfaces in order to preserve sharp features, such as edges or corners, on the mesh shapes. However, they often disregard the fact that such small weights on anisotropic surfaces still pose negative impacts on the denoising outcomes and detail preservation results on the shapes. In this paper, we propose a novel segmentation-driven mesh denoising method which performs region-wise denoising, and thus avoids the disturbance of anisotropic neighbour faces for better feature preservation results. Also, our backbone can be easily embedded into commonly-used mesh denoising frameworks. Extensive experiments have demonstrated that our method can enhance the denoising results on a wide range of synthetic and real mesh models, both quantitatively and visually.