Sparse Regularization-Based Approach for Point Cloud Denoising and Sharp Features Enhancement.

Area of Science:

• Computer Vision
• Geometric Modeling
• Computational Geometry

Background:

• Point cloud denoising is crucial for accurate 3D surface reconstruction.
• Existing methods face challenges in balancing noise reduction with sharp feature preservation.
• Noise and outliers in 3D scanning data degrade surface quality.

Purpose of the Study:

• To develop a robust point cloud denoising model.
• To preserve sharp geometric features during the smoothing process.
• To enhance the quality of reconstructed 3D surfaces.

Main Methods:

• A novel model combining L1-median filtering with sparse L1 regularization.
• Utilizing L1-median filter for robust outlier and noise handling.
• Applying L1 regularization to promote sparsity of sharp features.
• Optimizing the L1 minimization problem via proximal gradient descent.

Main Results:

• The proposed method effectively filters high levels of noise from 3D models.
• Sharp geometric features of the point clouds are successfully preserved.
• The denoising approach demonstrates comparable performance to state-of-the-art methods.

Conclusions:

• The L1-median filtering and sparse L1 regularization model offers an effective solution for point cloud denoising.
• This method enhances 3D surface reconstruction by maintaining critical geometric details.
• The approach provides a robust way to clean noisy 3D scan data while preserving essential features.