The Level-Set Segmentation Library (LSSEG)

1.0

With Anisotropic Smoothing Extension

Written by Odd Andersen

What is it? What does it do?

The LSSEG library was written during 2005/2006 in the context of a shared internal study project on PDE-base image processing, between the two SINTEF ICT departments Applied Mathematics and Optical Measurement Systems and Data Analysis.

The main two applications considered was anisotropic image smoothing and automatic, level-set based image segmentation . The LSSEG library contains algorithms and data structures that are useful when working in these contexts, and it also contains a set of sample programs where it demonstrates what can be done using the tools it provides.

How to read this documentation

This documentation is put together for the sole purpose of making it easier for the unfamiliar programmer to start using the contents of this library. It is not meant to be used as a theoretical introduction to PDE-based image processing. We have, however, provided links to a few hand-picked publications that explain the underlying theory of what we have done. We benefitted a lot from these publications when putting together this library, and want to express our thanks to the respective authors.

In order to get the general "feel" of this software's capabilities, we would suggest:

• read the pages referred to by How Segmentation in LSSEG Works and How Anisotropic Smoothing in LSSEG Works.
• compile and run some of the provided sample programs, then have a look at their code and see how they make use of the library.
• study the documentation of the main classes and algorithms, in particular the Image class, and (for segmentation) the LevelSetFunction and ForceGenerator classes. Some relevant algorithms would be anisotropic_smoothing(), compute_structure_tensor_XD(), compute_smoothing_geometry_XD(), LIC_XD_FS() (related to smoothing) and normal_direction_flow_XD(), develop_single_region_XD() and develop_multiregion_2D() (related to level-set functions and image segmentation).
• Have a good look at the referenced publications, particularly those chapters and sections mentioned elsewhere in this documentation. Nothing helps more than having a good theoretical understanding about what is going on.

Note:

This documentation is not written to be of a quality comparable to a scholarly paper. The list of references is not extensive. Rather than aiming to present the works in which various ideas were first developed, its purpose is to provide the reader with a limited reading list that would allow him/her to get a good grasp of the main underlying theoretical ideas upon which this software has been developed.

How Segmentation in LSSEG Works

We have written an introduction on the principles of segmentation in LSSEG, explaining the main concepts and algorithms. You can find it here.

How Anisotropic Smoothing in LSSEG Works

We have also written an introduction to the algorithm used for anisotropic smoothing in LSSEG. It is to be read here.

Sample programs

The sample programs found on this page make use of the library, and may be a natural starting point for exploring its capabilities and structure.

Other helpful parts of this documentation

• We have assembled a small glossary of terms used in this documentation.

• A page of references to works consulted can be found here.

---