Mesh Denoising Guided by Patch Normal Co-Filtering via Kernel Low-Rank Recovery

Area of Science:

• Digital Geometry Processing
• Computer Graphics
• Computational Geometry

Background:

• Mesh denoising is a critical challenge in digital geometry processing, aiming to remove noise without compromising essential surface features like sharp edges and fine details.
• Existing methods often struggle to balance noise reduction with the preservation of intrinsic surface properties.

Purpose of the Study:

• To develop a novel mesh denoising technique that effectively removes noise while preserving surface accuracy and intricate details.
• To address the limitations of current methods in handling varying noise levels and preserving surface features.

Main Methods:

• A new patch normal co-filter (PcFilter) is proposed, inspired by geometric statistics of similar surface patches.
• The method models mesh denoising as a low-rank matrix recovery problem, incorporating similar-patch collaboration.
• The model is generalized to kernel space for nonlinear data structures and optimized using block coordinate descent and proximal methods.

Main Results:

• The proposed PcFilter demonstrates competitive performance against state-of-the-art methods in both quantitative and qualitative evaluations.
• Results on synthetic and real-world data show significant improvements in surface accuracy and noise robustness.
• The method effectively preserves both shallow details and sharp features during the denoising process.

Conclusions:

• The PcFilter offers an effective and robust solution for mesh denoising, outperforming existing methods.
• The approach successfully balances noise removal with the preservation of critical surface geometric properties.
• This work contributes a valuable tool for digital geometry processing applications requiring high-fidelity mesh denoising.