Feature Preserving Mesh Denoising Based on Graph Spectral Processing.

Area of Science:

• Computer Graphics
• Computational Geometry
• Signal Processing

Background:

• 3D scanning generates noisy models requiring noise reduction techniques.
• Real-time applications demand high-quality 3D data, necessitating feature preservation.
• Variable environmental factors introduce complex noise patterns in 3D datasets.

Purpose of the Study:

• To introduce a novel coarse-to-fine graph spectral processing approach for 3D model denoising.
• To effectively remove noise from large-scale 3D scenes while preserving sharp geometric features.
• To address challenges posed by complex and varied noise characteristics in 3D surface data.

Main Methods:

• A coarse-to-fine graph spectral processing framework is proposed.
• A model-based Bayesian learning method identifies noise levels and feature subspaces in partial mesh segments.
• An iterative feature-aware fine step smooths face normals and vertices, preserving geometric details.

Main Results:

• The proposed approach demonstrates superior performance in reconstruction quality compared to state-of-the-art methods.
• Evaluations under diverse complex noise patterns confirm the method's effectiveness.
• The approach achieves a favorable balance between reconstruction quality and computational complexity.

Conclusions:

• The novel graph spectral processing method offers an effective solution for denoising noisy 3D models.
• Preservation of sharp geometric features is achieved even with complex noise patterns.
• The method provides a significant advancement for real-time 3D applications requiring reliable geometry generation.