SegAnIm: Segmentation and model based analysis of images of microstructures
Project Topic
Development of new, model based methods for segmentation and analysis of large 3d images of microstructures.
Project Description
Currently, in materials science there is a strong tendency towards
highly proprous and complex structures on smaller and smaller
scales. At the same time, new imaging techniques provide larger and
larger 3d image data at increasing resolutions. This development
causes new demands on segmentation and analysis of such micro
structures.
3d images of microstructures are of interest in various fields of
application, e.g. for quantitative structural analysis of materials,
for assessing materials properties or to get new insight into
production processes, but also in biology or medicine, e.g. in order
to distinguish between healthy and diseased tissue.
The analysis of new materials like highly porous foams or high
performance ceramics causes various problems. In this project, the
following very urgent problems (from the application point of view) are
addressed:
Segmentation of volume images
The concept segmentation is used for
two tasks -- finding the phase or image segment of interest in a
grey value image and identifying connected objects or regions.
Segmentation is crucial in image processing, as most analysis
methods work on segmented images, only. However, segmentation is an
ill-posed-problem, and often segmentation results are validated just
visually. This causes particular problems in dimensions higher than
two, where visual inspection can be misleading. Moreover,
segemtation methods for higher dimensional image data have to be
very efficient due to the amount of data. Work in this project
concentrates on
- efficient smoothing and segmentation of volume images of complex structures
- segmentation of fibre and edge systems based on locally adaptive anisotropic filtering and orientation analysis
Analysis of random networks
The skeletons of the strut system of open foams as well as fibre
systems can be modeled as random networks. Integral geometric
methods yield geometric characteristics for both the skeleton and
the original structure. The results depend on the structure, the
skeletonisation applied, as well as image resolution. A systematic
investigation is needed to find out which analysis method should be
prefered under which requirements.
Models and model fitting for cellular structures
Geometric models for cellular structures are needed because of a
lack of 3d images due to insufficient resolution (e.g. for ceramics)
and in order to systematically study the relations between
structural and macroscopic materials properties (e.g. for
foams). Random tessellations are a well suited class of
models. However, model fitting is a difficult and tedious task as
most realistic models can not be described analytically.
Project Members
Project Chair
Participating Research Groups
- Sect. Models and Algorithms in Image Processing, Fraunhofer Institute for Industrial Mathematics (ITWM)
- Image Understanding and Pattern Recognition Group, Department of Computer Science
- Applied Mathematical Statistics Group, Department of Mathematics
Scientific Personnel
- Oliver Wirjadi (Sect. Models and Algorithms in Image Processing and Image Understanding and Pattern Recognition Group)
- Dr. Claudia Lautensack (currently Chalmers Teknikska Högskola)
- Tetyana Sych (Sect. Models and Algorithms in Image Processing and Applied Mathematical Statistics Group)
External Cooperation
- Dr. Aila Särkkä, Chalmers Teknikska Högskola, Göteborg
Project Events and Achievements
- Project start date: November 1st, 2005
- Project end date: December 31st, 2007
Workshops
- Workshop 3d Image Analysis and Modeling of Microstructures, 25/26 April 2007, Fraunhofer ITWM
Project Publications
Christoph H. Lampert, Oliver Wirjadi. In:
IEEE Trans. Image Processing. Volume 15, Number 11, P. 3501--3513, November, 2006
Katja Schladitz, J. Ohser, W. Nagel. In: A. Kuba and L. G. Nyul and K. Palagyi ed.,
13th International Conference on Discrete Geometry for Computer Imagery, Szeged, Hungary. LNCS DGCI, Springer, P. 247--258, 2006
Claudia Lautensack, Katja Schladitz, Aila Särkkä. In:
Proceedings of the 6th International Conference on Stereology, Spatial Statistics and Stochastic Geometry, Prague. 2006
Oliver Wirjadi, Thomas M. Breuel, W. Feiden, Yoo-Jin Kim. In:
Bildverarbeitung für die Medizin. Informatik aktuell, Springer, P. 76--80, 2006
Claudia Lautensack, T. Sych. In:
Image Analysis and Stereology. Volume 25, P. 87--93, 2006
