Research Projects

Introduction

Research interests in single-modality nonlinear registration include four different kinds of subproblems. Deformable registration of two or more CT lung data sets at different states in the breathing cycle going from Functional Residual Capacity (FRC, expiration) to Total Lung Capacity (TLC, inspiration) for modelling breathing motion and deriving lung ventilation. Deformable registration of a contrast-enhanced and a native CT lung data set for deriving lung perfusion. Deformable registration of contrast-enhanced and native CT liver data sets at one or several phases in the contrast-uptake cycle for liver perfusion. And finally, highly accurate partially rigid bone registration for head and neck CT-Angiography applications to extract bone structures from CTA images.

Deformable Lung Registration

The input for this task consists of native CT thorax scans at two or more different breathing states between Total Lung Capacity (TLC, inspiration) and Functional Residual Capacity (FRC, expiration). Deformable registration of distinct breathing states is a prerequisite for deriving ventilation information by simple subtraction of expiration from inspiration data or by fusion with special functional scans and it leads to models of breathing motion in the lung. For this purpose we have available high resolution sheep lung data at up to five distinct static breathing states and human lung data at inspiration/expiration. Another application of deformable lung registration is the fusion of native and contrast-enhanced CT lung data to show perfusion information again either by subtraction or by fusion with a special scan. A notion of vessel consistency should be included in the deformable registration, since it is important that the same amount of vessels is regarded before and after registration.

Deformable Liver Registration

Similar to the lung registration, liver registration for perfusion measurements is a topic of interest. Contrast-enhancing techniques are used to get up to 8 liver images at different phases of the contrast uptake cycle. Each of these images has to be registered to a native scan to correct motions due to breathing. Afterwards subtraction techniques are used to derive the amount of perfusion in the liver. The setup of the registration algorithm is very similar to the lung registration problem.

Partially Rigid Bone Registration

The intended application of rigid bone registration is a very accurate registration of bones from native and contrast-enhanced CT images of the head and the neck. In contrast-enhanced images vessels and bones have very similar intensities, such that simple segmentation algorithms like thresholding do not work which are frequently used for CTA image studies. The intended strategy for the removal of bone structures is to take a simple (threshold-based) bone segmentation taken from the native image and register it to the contrast-enhanced image. Registration is necessarysince small patient movements may occur (especially in the neck and shoulder area) between the acquisition of both kinds of images. Registration has to be very accurate in this area, since there are vessel structures that lie close to or inside the bone structures as well. It can be assumed that the bones themselves are rigid but the relative position of bones to each other may change. Pairs of bones should be registered rigidly but the relative bone movements are taken into account leading to a partially rigid registration scheme.

When a dust laden gas is sucked through a filter, the dust remains on its surface and forms a compact dust layer called filter cake. Periodically the filter cake is at least partially removed by inverse high pressure air pulses to allow continuous operation of the filter. Knowledge of the distribution of the filter cake on the filter surface at different stages of operation is decisive for filter operation. It is shown that the principle of calibrated shape from stereo with pattern projection for generating texture on the surface gives a robust 3D reconstruction of the filter surface when accessing through a glass window. Rigid registration of surface patches using landmark points, combined with an Iterative Closest Point Algorithm as a refinement procedure gives a continuous 3D model of the entire visible filter surface. Cake thickness is calculated by taking the height difference of two surface models, acquired before and after dust deposition. The challenging problem is to account for the non-rigid deformation of the filter cloth. A global deformation model is estimated using Thin Plate Spline Interpolation based on landmark points on the filter surface.

(Matlab) - collection of some most promising methods and algorithms related to appearance based object recognition.

Most technological paper qualities are directly or indirectly influenced by the three-dimensional paper structure which means the spatial arrangement of fibres, fillers, pores and if necessary of the coating. In the area of the structure analysis, two main groups can be distinguished: destructive and non-destructive approaches.

The paper structure analysis methods have to fulfil the following basic requirements. The quality of the 3D data must be high enough to make an exact delimitation between the relevant paper contents possible. Furthermore the spatial resolution must at least lie in the area of a micron. In order to be able to make statistically reasonable statements for the complete paper sample, sizes of the range of at least one square centimetre must be analyzed. To conclude, the time expenditure to the digitalization of the paper sample should also lie within a practicable area of hours.

Primary objective of the project is the detailed analysis of the fibre network structure of a paper sample by means of image analysis methods. Still this application is an extremely complex problem and in this resolution (microns) and sample size (square centimetre) unsolved. Therefore the development of a completely new concept is necessary. This concept should finally make a general assignment of the three-dimensional paper structure to pre-defined paper-ingredients classes possible. As a final result the segmentation of single fibres and their three-dimensional tracking is realizable.

Soft tissue like tendons, arteries, veins or skins are important biological materials. A greater understanding of the foundations and interactions of structure and function of soft tissue, and, in particular, the associated mechanobiology is of great interest in the field. A thorough understanding of the complex interrelations between mechanical factors and the associated biological responses may help to improve diagnostics which allow disease and injury to be treated earlier. The research proposed here will develop a fully automatic system for analyzing macroscopic structures obtained from histological images of arteries by means of modern computer vision techniques. Besides being interesting from the mechanobiological point of view the structural analysis of images of collagen fibers poses also several challenging questions from a computer vision point of view. In particular, due to the wide variety of different appearances of collagen fibers in images this task is non trivial. The main task of this research is the development of novel segmentation techniques for robustly segmenting individual fibril bundles and estimating their parameters, like location and shape, fibril density, mean fibril orientation, wriggling of fibrils etc. This will be achieved by developing novel perceptual grouping methods operating on the extracted orientation data of fibrils. Another major challenge of this research is to extend the structural analysis from 2D to 3D.

Current paper and corresponding presentation

The concept of "directed evolution" is dependent on the availability of systems that allow the identification of interesting enzyme variants within a large, artificially generated diversity. Thus, one prerequisite is the availability of systems that allow the detection of specific enzyme features such as activity, selectivity, stability etc. at smallest culture scales and at high-throughput conditions. Therefore, focus is put on the development of methods witch are rapid, sensitive and cost-effective and preferably work at the microwell-plate, single colony or even single cell level.

Surrogate substrate analogues which would allow easy detection of reaction products by e.g. fluorescence techniques are not feasible due to the "First Law of Directed Evolution" which says "you get what you screen for". The "real" substrates should be converted or at least derivatives that are very close to these substrates have to be used in such systems. Therefore, general analytical methods which allow following the enzyme-catalyzed reaction of any desired substrate are developed.

Another important prerequisite is the availability of methods that allow a high degree of automation. Such methods include high throughput detection systems based on image analysis and software for accurately recognizing hits. In addition, statistical methods and systems for data management are developed to properly set up and evaluate the results of screening programs and to handle large data volumes.

Robust and Adaptive Approaches to Scene and Object Recognition: The goal of this joint project is to investigate new robust and adaptive approaches in the area of object and scene recognition. Object and scene recognition is a necessary requirement for developing truly cognitive systems as well as for the development of advanced and novel multimodal interfaces leading to ambient intelligence. Having a robust object and scene recognition system the following applications will greatly benefit: novel user interfaces which understand human activities, intelligent surveillance, indexing multi-media databases and content analysis of images, autonomous mobile systems and robotics, industrial inspection and robotics, etc. The goal is to develop computer vision based systems that can recognize objects, and in the context of environment perform localization and navigation. The major challenge is to develop systems and methods that can work under realistic unconstrained conditions (i.e., outside the lab). The three partners proposing this project (Center for Machine Perception, Czech Technical University Prague, CMP, Computer Vision Lab, Faculty of Computer and Information Science, University of Ljubljana, CVL, and Institute for Computer Graphics and Vision, Graz University of Technology ICG) have considerable expertise in this area and developed complementary methods and techniques. The goal of the project is to join the efforts and combine the expertise. In particular, we do the following activities:

• Organization of joint workshops and colloquia
• Exchange of PhD. students
• Joint research/joint papers

The aim of this project is the development of an imaging system for an industrial partner. The developed system should survey entrances using video images and register the people passing the entrance. The system has to operate under outdoor conditions (sun light, fog, etc.).

The ultimate limit of today‘s user interfaces lies in the used two-dimensional abstractions that are only effective for their original domain of document-centric work. In contrast, in our everyday world we are not limited to our desktop surface. Information in the real world is perceived and processed in three dimensions, continuously and in real time. A human-computer interface that can capture these properties will be able to render a new level of services to the user, enabling the use of computers for new application domains and for new user populations. This „anywhere“ and „anytime“ requirement for pervasive computing cannot be fulfilled with miniature versions of desktop environments. A new style of user interface, a paradigm shift is needed. Therefore, the core of the proposed research is the following thesis: Augmented reality (AR), i. e., enhancing a user‘s perception of the real world with computer generated graphics and annotations, can make working with computers in 3D as productive as the desktop metaphor in 2D. This thesis is motivated by the fact that AR allows to integrate the whole world into the interface – the world essentially becomes the interface. Therefore users are able to leave their physical desktops and computer desktops to interact with their environment and with other users. The AR platform Studierstube lead by the proposer is world-wide unique in its combination of augmented reality, 3D display and groupware elements. Studierstube, the study room where Faust was searching for enlightenment, describes the philosphy of using the place as a mediator to information and insight. We are currently developing a wearable augmented reality system, which allows the user interface to be in any place, with and for anybody. Within the proposed augmented reality/pervasive computing infrastructure, an environment can be turned into a virtual „ether“ encompassing users that are enabled to interact with the computer through realworld objects. The proposed project work will expand the augmented reality platform Studierstube into a pervasive computing environment built on a variety emerging technologies. A number of promising application areas is selected, for which applications will be developed that try out the new style of interfacing with the computer in practice. Influence

The scanning electron microscope (SEM) is an important tool to examine very small structures. Its large magnification combined with good contrast and large depth of view make it possible to view and characterize microscopic structures in the sub-micron scale. In the recent years, the problem of dense surface reconstruction from multiple SEM images was a research topic on this institute. Reconstruction approaches like shape from stereo and shape from photometric stereo have been evaluated. This work presents a framework for automatic scene reconstruction from three images acquired by a scanning electron microscope. The basic assumption is that the specimen is tilted eucentrically in front of the camera, camera geometry is assumed to be unknown but constant over all views. It is shown that methods for estimating the trifocal tensor as well as modern auto-calibration approaches can be adapted to the imaging conditions in the SEM, and Euclidean scene structure can be retrieved from three uncalibrated views. The performance of the proposed framework is evaluated on synthetic data as well as real images. It is shown that Euclidean scene structure can be retrieved robustly under varying image noise and inaccurate initialization.