Summary of research

According to the founding documents, the objective of the CBA is ``to create the know-how needed for an operative and sensible use of digital image analysis in society, particularly in the fields of environment, medicine, and industry.'' The research work is more or less organised in three groups. The borders between the groups are becoming more and more fuzzy, as many projects enclose people from several groups. The groups are: The Image analysis group at UU, headed by Ewert Bengtsson, which works mainly with medical applications; the Aquatic remote sensing group at UU, headed by Tommy Lindell; and the Image analysis group at SLU headed by Gunilla Borgefors, which concentrates on forestry applications and digital geometry, but also conducts research on medical applications. The order of groups and projects below is arbitrary.

The SLU Image Analysis group has the aim to be a central SLU source for image analysis knowledge. This means that we conduct basic image research, in addition to being involved in a number of applications, from forest inventory via paper fibre analysis in situ to virology. The main problem in the group in the past has been the lack of seniors, except the professor. However, from this year we employ an Assistant Professor (``Forskarassistent'') for the first time.

For a number of years we have been working on wood fibre applications. One project investigates the fibre network in paper. The ultimate goal is to understand how individual fibres build up the paper and what effect different types of fibre networks have on paper properties. The necessary first stage, to actually produce a high-resolution volume (3D) image of paper, which is quite difficult in itself, was accomplished about two years ago. Now, we have worked on shape analysis, at the individual fibre level, the fibre network level, and at the paper level, and developed algorithms for computing a number of useful measures. This project is part of the national VISIT programme, funded by SSF, and is done in co-operation with StoraEnso Research, Falun and Dept. of Science and Engineering, Linköping University, Campus Norrköping. Mattias Aronsson successfully defended his PhD thesis in this project 12 Dec. 2002.

Another fibre project, that was finished this year developed methods to automatically analyse fibre morphology in confocal microscopy transverse sections of wood. For the forest industry to be able to maximise the use of the wood fibre potential, more knowledge of the fibre morphology is needed. >From 1998, the work has been done in close collaboration with Forest Research New Zealand Ltd, and this co-operation will continue, as Mattias Moëll, who become a Doctor in this project Dec. 2001 is now employed at this company.

Two other small projects in Bengtsson's group deals with fibre and paper analysis.

Forest inventory from air-borne sensors have been an active and productive research field in the group since its beginning in 1994. The aim is to make inventory from such data so detailed and correct that it can replace field inventories, except for small investigations to collect ground truth. At present PhD student Mats Erikson is working on very accurate segmentation methods for tree crowns, and species identification based on this segmentation. The goal is to be able to differentiate between spruce and pine. Even though they have the same spectral signatures, they do have, on the average, different shapes and internal structure. This year, we have also worked together with Dept. of Forest Resource Management and Geomatics, SLU, Umeå to evaluate tree segmentation algorithms.

We also are involved in a pilot project to evaluate if horizontal laser scanning data is suitable for guiding a robot for thinning young forests.

A Post-Doc. stipendiate, Lucia Ballerini, has been working together with Dept. of Food Science, SLU in the FOOD21 programme funded by MISTRA. Mostly, the work has been on composition and quality of meat, using colour photographs and magnetic resonance images. Most of the work this year concentrated on evaluation of different meat processing methods.

The SLU group also works on several medical applications (a consequence of the CBA environment).

PhD student Xavier Tizon investigates the uses of grey-level connectivity and fuzzy set theory in 2D and 3D images. The main application is arteries-veins separation in magnetic resonance angiography images. Spin-offs also investigate segmentation of other images using the same principles, e.g., detection of liver cancer metastases. We co-operate with Dept. of Medicine and Care, Linköping University. This project, is part of the national VISIT programme, and involves both groups at CBA.

A new project, where Xavier Tizon is our main researcher has been started during the year. The task is to develop measures for and measure the total plaque burden index in whole-body magnetic resonance images. If this is successful, it can lead to massive scanning programmes for early detection of arteriosclerosis. This work is done in close co-operation with Dept. of Oncology, Radiology, and Clinical Immunology, UU Hospital.

We also investigate magnetic resonance angiography (MRA) images using our latest theoretical tools. The 3D overall tree structure and the thickness of the blood vessels is modelled using curve skeletons (see below).

PhD student Ida-Maria Sintorn is involved in two virus applications. The first is the identification and classification of human cytomegalo virus. Available transmission electron microscopy images of infected cell nuclei are quite noisy. We have reached quite good results using normalised mean radial grey-tone profiles for the three main degrees of virus maturation. This work is done together with Centre for Molecular Medicine, Karolinska Hospital, Stockholm. Another virus application is to investigate the 3D shape of HIV viruses by reconstruction from a series of electron micrograph projections. Ingela Nyström is also active in both these projects.

In the theoretic work, the group concentrated on digital geometry in 3D and on extending methods for binary image processing to fuzzy images, to avoid the difficult and data-destroying binarisation step.

An old ``obsession" of the group is the development of digital distance transforms. They are very useful tools for many types of image analysis. This year, the work included developing more exact distance transformations in 3D, using neighbourhoods up to 5 × 5 × 5 voxels. In medical and industrial volume images, the picture elements are often rectangular boxes rather than cubes. It is preferable to work directly in such grids, rather than interpolating the image to a cubic grid. We have optimised distance transforms in such grids. We have also investigated distance transforms in 4D, using 3 × 3 × 3 × 3 neighbourhoods. Finally, we have developed a way to apply distance transforms to fuzzy segmented objects. The various people involved in distance transform research are Gunilla Borgefors, Ida-Maria Sintorn, and Stina Svensson.

Assistant Professor Stina Svensson continues to work on digital shape in volume images. One direction of the research is the decomposition of 3D objects into significant parts, i.e., nearly convex parts and elongated parts, using distance transforms. It is often necessary to apply similar processes to the background of the objects and to compute and analyse their convex deficiencies. Another research direction is skeletonization (or thinning), which is the process where objects are reduced to structures of lower dimension. Objects in 3D can be reduced to surfaces or even to curves. This year, we have started serious work on skeletonization directly on grey-level images. One application is finding central lines through blood vessels in MR images. The computations of curve skeletons have also led to work on analysis of digital curves in 3D. Much of Svensson's work is done in co-operation with Istituto di Cibernetica, Napoli, Italy. At CBA, Ingela Nyström and Ida-Maria Sintorn are heavily involved in several aspects of this work. Finally, in co-operation with the Pattern Recognition Section at Delft University of Technology Svensson has started work on shape analysis in 4D.

PhD student Nataša Sladoje is working on the new project: Fuzzy shape analysis in 2D and 3D. This means development of shape analysis directly in grey-level images or in fuzzy segmented images. So far, the work has been concentrated on accurate and precise measurements of fuzzy objects that have low resolution. Ingela Nyström is assistant supervisor for Nataša Sladoje.

Another small grey-level shape project that started this year regards the computation of grey-level convex hulls.

Shape description derived from volume images is usually local, which prevents comparison between and analysis of shapes in different translations and rotations. We aim to develop global shape descriptions using a linear combination of spherical harmonics. The problem has been solved for starshaped objects and ``nice" non-starshaped ones, but not yet generally. PhD student Ola Weistrand is working on this, in close co-operation with the Dept. of Mathematics at UU.

Lucia Ballerini has been working on genetic snakes, i.e., active contour models optimised using genetic algorithms.

The UU Image Analysis group has always had its main focus on medical applications of image analysis and visualisation, where tomographic volume images from different sensors and light microscopic images of tissues and cells have been analysed.

Earlier, our work on light microscopy images has mainly focused on absorption images. A few years ago, we started working also on fluorescent light microscopy images. This has involved two different projects. Together with researchers at the Karolinska Institute, we have developed a new way of extracting information from multiple stainings of the same cells. Together with the company Amersham Biosciences we have developed new methods for automatic cell image segmentation, including automatic splitting and merging to handle overlapping cells as well as new methods for feature extraction, in particular the quantitative detection of so called ``ruffling" cells. Carolina Wählby and Joakim Lindblad are the two PhD students working on these projects. Towards the end of the year a third student, Patrick Karlsson was recruited to this group.

Continuing our work on 3D tomographic images PhD student Roger Hult has worked in the Hubin (Human brain informatics) project at the Karolinska Institute and Hospital where he has implemented and tested his brain segmentation methods for MRI images. The main application is the relation between brain morphology and psychiatric conditions.

Another PhD student, Seyed Rahman Razifar, is working at the PET Centre, UU where he is studying how the multivariate information obtained from PET studies can be analysed and presented together with the multi-dimensional image data (3D+time). This year the work has mainly been focused on modelling the PET image formation processes.

PhD student Felix Wehrmann is working theoretically, and has continued his studies of fundamental methods for expressing the variation of shapes of natural objects in images. The most promising approach has turned out to be based on neural networks.

Another basic methods development project deals with the problem of representing and accurately and precisely measuring surface area and volume of 3D digitised objects. Especially measuring the surface is a difficult problem. This is being done mainly by Ingela Nyström and Joakim Lindblad in collaboration with Jayaram K. Udupa, University of Pennsylvania.

In a computer graphics project, Bengtsson is supervising PhD student Anders Hast from Gävle University College, who, together with Tony Barrera at Cycore Inc., has been studying how various image rendering algorithms can be implemented more efficiently. Several basic algorithms for shading and bump mapping has been significantly improved.

Fredrik Bergholm has continued his research on plenoscopy, i.e., a new way of optically capturing and encoding 3D information in single 2D images.

In support of the different projects at CBA, as well as in order to provide an improved platform for the education in image analysis, we have for many years worked on our general platform for image analysis, the IMP system. Towards the end of the year we decided to reorganise the many changes and additions to the system that has been implemented over the last several years into a completely new system. This is expected to be completed in mid 2003.

The research of the UU Aquatic Remote Sensing group is focused on different environmental applications of digital remote sensing. The present activities vary from mapping and monitoring of algae blooms and distribution of plumes to mapping and monitoring of tropical coasts and sea bottoms.

One important area of research is our continued development of image analysis methods for imaging spectrometry. Much effort has been put into the procedures for pre-processing of remote sensing data and the development of bio-optical modelling for more operational monitoring of water quality from space.

A big, national, project called RESE (REmote Sensing for the Environment) is focused on methods for detecting changes in aquatic ecosystems and monitoring of algae blooms. This project was built on and continued the work in the successful SALMON project, an EU project on monitoring water quality in European lakes, which was completed in 1999. The long-term goal here is using satellite, together with airborne hyperspectral data, for algae bloom detection, eutrophication, and pollution in Nordic waters. One aspect of the latter we have worked on is detection of industrial plumes in lakes and seas. Bio-optical modelling continues to be an important and necessary part of the work. At present PhD student Petra Philipson is working in the RESE project.

The co-operation with the Italian groups that participated in the SALMON project is continuing in ROSALMA, monitoring of chlorophyll and macrophytes from satellites, and NYMPHA, experimentation on a remote sensing integrated system for lake water monitoring.

Our work on the detection of coral bleaching from remote sensing sources has continued. It now includes sensors like IRS-LISS-III, SPOT, and IKONOS. Field work in Belize has been done during the year, to make evaluations possible. This project is also financed via RESE and Petra Philipson is the main researcher.

A new project in the group is focused on acquisition and colour correction of underwater multi- or hyperspectral data (e.g., colour photos). This can be important for many applications, such as marine biology and underwater archaeology. The research is carried out by PhD student Julia Åhlén at Gävle University College.

We also work theoretically on developing techniques for analysis of hyperspectral image data. This research in now mainly done by PhD student Hamed Hamid Mohammed. An important aspect is developing linear transformations method, based on such transforms as the familiar PCA (Principal Component Analysis) and the more recent ICA (Independent Component Analysis.) Hamid Mohammed also develops new techniques for information extraction using neuro-fuzzy systems, i.e., so-called Weighted Neural Networks (WNN).

Finally Hamed Hamid Mohammed has co-operated with Anders Larsolle at Dept. of Agricultural engineering, SLU, (Gunilla Borgefors is his assistant supervisor) on information extraction from hyperspectral crop reflectance data.

We have also supervised four Master Theses that were completed this year, most with industrial co-operation partners. They treated: Registration of electrophoresis gel images; Automatic high speed sorting of seed for quality control; Analysis of coal inclusions in cast iron; and Automatic detection and reading of road signs from a camera in a car window. Several other Master theses are in progress and will be finished in 2003. Most of these are also industrial co-operations, e.g., Automatic detection of damages to pantographs on locomotives using image analysis; Detection and automatic reading of number plates on cars; and A system for quality control of shiny surfaces.