Efficient surface reconstruction from noisy data using regularized membrane potentials.

Area of Science:

• Computer Vision
• Computer Graphics
• Computational Geometry

Background:

• Surface reconstruction from sparse and noisy data is a challenging problem.
• Existing methods often struggle with noise tolerance, outlier removal, and detail preservation.
• Orientation information is frequently required, limiting applicability.

Purpose of the Study:

• To develop a physically motivated surface reconstruction method.
• To improve noise tolerability and outlier removal while preserving details.
• To achieve smoother surfaces compared to existing techniques.

Main Methods:

• Utilizing regularized-membrane potentials for point aggregation and noise reduction.
• Employing an iterative algorithm for classifying grid points based on scalar field properties.
• Implementing a mass-spring system with bending-energy minimization for mesh smoothing.

Main Results:

• Successfully recovered smooth surfaces from noisy and sparse datasets.
• Demonstrated improved noise tolerability and outlier removal.
• Generated smoother surfaces than piecewise linear representations.
• Achieved favorable speed and flexibility compared to prior approaches.
• GPU acceleration enhanced computational efficiency.

Conclusions:

• The proposed method offers a robust and efficient solution for surface reconstruction.
• It effectively handles noisy and sparse data, preserving important features.
• The technique advances the state-of-the-art in surface reconstruction and image segmentation.