Laguerre approximation of 3D images
This software package provides an implementation of the algorithm proposed in . It aims to detect a Laguerre tessellation that approximates a given grain structure as well as possible. An interface-based discrepancy measure is minimized using the cross-entropy method, a robust stochastic optimization technique.
 A. Spettl, T. Brereton, Q. Duan, T. Werz, C.E. Krill III, D.P. Kroese and V. Schmidt, Fitting Laguerre tessellation approximations to tomographic image data. Philosophical Magazine 96 (2016), 166-189.
If you use this software package in a work that leads to a scientific paper, we ask you to mention it explicitly and cite reference . A preprint of the paper is available at arXiv:1508.01341 [cond-mat.mtrl-sci].
This project requires Java 7 or newer. To compute suitable generators of a Laguerre tessellation for a labeled 3D image "artificial_pcm_input.tif" (located in the subfolder "data"), run:
java -Xmx3500m -jar LaguerreApproximation3D.jar data/artificial_pcm_input.tif data/artificial_pcm_output.txt
In this example, the detected generators will be written to the text file "artificial_pcm_output.txt". A Java heap size of 3500 MB is used, which is necessary in order to process the example data set. The 3D image itself is a labeled image, i.e., label zero stands for background, label 1 for the first grain, label 2 for the second grain, etc. The grain regions should be roughly convex. Furthermore, it is expected that there is at most a thin layer of background voxels (1 voxel thickness) between adjacent grains. A multi-page image file is required as input, i.e., a single file contains several 2D images which correspond to the slices (in z-direction) of the 3D image. For example, ImageJ / Fiji can generate such 3D image files. The output text file contains the detected generators. One line corresponds to one generator. The first value is the label, the remaining four values are x-, y-, z-coordinates and the radius.
- The executable JAR file "LaguerreApproximation3D.jar" is contained only in the binary release. If you want to compile the source code yourself: the class "LaguerreApproximation3D" in the package "laguerre_approximation_3d" contains the main method.
- The experimental data set requires about 4500 MB Java heap (due to the cylindrical shape of the sample).
The following optional arguments may be used. Here, they are given with their default values:
--M=4000 --rho=0.05 --tauInject=10 --deltaInject=0.05 --injections=-1 --gamma=0.9 --kappa=0.25 --tauTerminate=10 --deltaTerminate=0.01
The parameters are explained in . The only exception is the
parameter, which allows you to choose the number of variance injections manually.
The default value "-1" stands for the standard behavior, i.e., automatic
selection of the number of variance injections. If a value equal to or greater
than zero is specified, infinity is assumed as a new default value for
(in order to not abort variance injections early).
Source and Javadoc
The source code is supplied in the folder "src" and it contains Javadoc comments. The main class is "laguerre_approximation_3d.LaguerreApproximation3D".
Copyright and licenses
- Copyright © 2015-2016 Institute of Stochastics, Ulm University
This software is licensed under the GNU General Public License v2.0 (published in June 1991), or (at your option) any later version. The full text of GPL 2 is available at https://www.gnu.org/licenses/gpl-2.0 and in the file "LICENSE".
- 2016-02-02 - Updated preprint, added publication details, no code change.
- 2015-08-06 - Initial release.