Bi-Normal Filtering for Mesh Denoising.

Area of Science:

• Computer Graphics
• Computational Geometry
• Digital Signal Processing

Background:

• Current mesh denoising methods often rely on either facet normals or vertex normals, potentially missing crucial geometric information.
• Utilizing only one type of normal field can lead to the overlooking of intricate surface details and features.

Purpose of the Study:

• To propose an effective mesh denoising framework that integrates both facet and vertex normal fields.
• To leverage the complementary information from both normal fields for enhanced denoising quality.

Main Methods:

• A novel framework integrating facet and vertex normal fields is proposed.
• The method involves vertex classification, bi-normal filtering using piecewise smooth region clustering, and a quadratic optimization algorithm for vertex position update.
• The core idea is to decompose inconsistent normal fields into piecewise consistent ones in challenging regions.

Main Results:

• The proposed bi-normal filtering and integration method achieves higher quality denoising results compared to existing approaches.
• Experimental results demonstrate superior performance on surfaces with complex geometric features and irregular sampling.
• The technique effectively preserves multifarious geometric details often lost in single-normal-field methods.

Conclusions:

• Integrating both facet and vertex normal fields offers a more comprehensive approach to mesh denoising.
• The proposed framework effectively handles challenging regions by exploiting piecewise normal field consistency.
• This method advances the state-of-the-art in 3D mesh surface denoising, particularly for complex models.