MATLAB code for the article "Implicit reconstructions of thin leaf surfaces from large, noisy point clouds".
If you use this code in your work, please cite the following paper:
Riley M. Whebell, Timothy J. Moroney, Ian W. Turner, Ravindra Pethiyagoda, Scott W. McCue, Implicit reconstructions of thin leaf surfaces from large, noisy point clouds, Applied Mathematical Modelling, Volume 98, 2021, Pages 416-434, ISSN 0307-904X, https://doi.org/10.1016/j.apm.2021.05.014.
This source code is dependent on the following MATLAB toolboxes:
- Statistics and Machine Learning Toolbox,
- Partial Differential Equation Toolbox, and
- Computer Vision Toolbox.
The main parameters of interest and their effects are:
- Downsampling grid size:
downsampleParam. Higher values will reduce runtime but also reduce reconstruction resolution. - Smoothing parameter:
rho. A higher value will yield a smoother (as defined by the thin plate penalty) surface reconstruction. - Sampling resolution:
Ngridfor marching cubes, orHmaxfor marching tetrahedra.- A higher value for
Ngridwill result in a finer grid for marching cubes isosurfacing. - A lower value for
Hmaxwill result in a finer tetrahedral mesh for marching tetrahedra isosurfacing.
- A higher value for
sphereTest.mwill reconstruct one half of a sphere from 791 points, without denoising or downsampling. This should run in under a minute on a desktop computer.main.mwill reconstruct a plant leaf from a heavily downsampled point cloud with 5516 points. This should run in under 10 minutes on a desktop computer. You may like to reduce thedownsampleParam(say, to 0.5) for a finer reconstruction.