Volker Metzler

Illumination-invariant change detection

Moving objects in image sequences acquired by a static camera can be detected by analyzing the grey-level difference between successive frames. Direct motion detection, however, will also detect fast variations of scene illumination. This paper describes a method for motion detection that is considerably less sensitive to time-varying illumination. It is based on combining a motion detection algorithm with an homomorphic filter which effectively suppresses variable scene illumination. To this end, the acquired image sequence is modelled as being generated by an illumination and a reflectance component that are approximately separated by the filter. Detection of changes in the reflectance component is directly related to scene changes, i.e., object motion. Real video data are used to illustrate the systems performance.

Defect interpolation in digital radiography: how object-oriented transform coding helps

Todays solid state flat panel radiography detectors provide images which contain artifacts caused by lines, columns and clusters of inactive pixels. If not too large, such defects can be filled by interpolation algorithms which usually work in the spatial domain. This paper describes an alternative spectral domain approach to defect interpolation. The acquired radiograph is modeled as the undistorted image multiplied by a known binary defect window. The window effect is then removed by deconvolving the window spectrum from the spectrum of the observed, distorted radiograph. The basic ingredient of our interpolation algorithm is an earlier approach to block transform coding of arbitrarily shaped image segments, that extrapolates the segment internal intensities over a block into which the segment is embedded. For defect interpolation, the arbitrarily shaped segment is formed by a local image region with defects, thus turning extrapolation into defect interpolation. Our algorithm reconstructs both oriented structures and noise- like information in a natural-looking manner, even for large defects. Moreover, our concept can also be applied to non- binary defect windows, e.g. for gain correction.

Scale-independent shape analysis for quantitative cytology using mathematical morphology

A system for automatic quantification of morphological changes of cell lines, proposed for cytotoxicity tests of biomaterials, is presented. Light-micrographs of cultured cells are segmented by adaptive thresholding within a local adaptive window. Connected cells in binarized micrographs are separated by a novel morphological multiscale method, treating cells in their size-specific scale and hence resulting in scale-independent separations. Significant shape descriptors correlating well with cell toxicity are extracted from single cells. Size and compactness distributions turned out to be reliable and useful parameters, providing an alternative to the common subjective grading of shape deformations by visual inspection. The system is evaluated for several standardized toxical reference substances and is now in use for clinical biocompatibility testing.

Perfusion harmonic imaging of the human brain

The fast visualisation of cerebral microcirculation supports diagnosis of acute cerebrovascular diseases. However, the commonly used CT/MRI-based methods are time consuming and, moreover, costly. Therefore we propose an alternative approach to brain perfusion imaging by means of ultrasonography. In spite of the low signal/noise-ratio of transcranial ultrasound and the high impedance of the skull, flow images of cerebral blood flow can be derived by capturing the kinetics of appropriate contrast agents by harmonic ultrasound image sequences. In this paper we propose three different methods for human brain perfusion imaging, each of which yielding flow images indicating the status of the patients cerebral microcirculation by visualising local flow parameters. Bolus harmonic imaging (BHI) displays the flow kinetics of bolus injections, while replenishment (RHI) and diminution harmonic imaging (DHI) compute flow characteristics from contrast agent continuous infusions. RHI measures the contrast agents kinetics in the influx phase and DHI displays the diminution kinetics of the contrast agent acquired from the decay phase. In clinical studies, BHI- and RHI-parameter images were found to represent comprehensive and reproducible distributions of physiological cerebral blood flow. For DHI it is shown, that bubble destruction and hence perfusion phenomena principally can be displayed. Generally, perfusion harmonic imaging enables reliable and fast bedside imaging of human brain perfusion. Due to its cost efficiency it complements cerebrovascular diagnostics by established CT/MRI-based methods.

A novel object-oriented approach to image analysis and retrieval

Common image processing tasks such as quantitative analysis, classification, or image retrieval require content-based techniques to firstly detect visually perceivable structures that have a semantic interpretation for a specific observer in a certain context and secondly to describe their properties in a comprehensive way. To achieve these aims, we propose an object-oriented approach to image interpretation utilizing a morphological multiscale decomposition to transform an image into a hierarchical data structure that represents image objects by their topological relations and descriptive attributes. The object hierarchy can be stored in a relational image archive and serves as interface to a rule-based expert system that either evaluates image objects directly or compares them with those of the stored images. Thus, both image analysis and retrieval can be realized by appropriate queries to the expert system. The system has already been used successfully for quantitative analysis and classification of biomedical and aerial images.

Segmentation of medical images by feature tracing in a self-dual morphological scale-space

The multiscale approach derives a segmentation from the evolution of appropriate signal-descriptive features in scale-space. Features that are stable for a wide range of scales are assumed to belong to visually sensible regions. To compensate the well-known drawbacks of linear scale- spaces, the shape-preserving properties of morphological scale-space filtering are utilized. The limiting duality of morphological filters is overcome by a selfdual morphological approach considering both light and dark structures in either the opening or the closing branch of the scale-space. Reconstructive opening/closing-filters enable the scale=analysis of 2D signals, since they are causal with respect to regional maxima/minima. This allows to identify important regions in scale=space via their extrema. Each extremum is assigned a region by a gradient watershed of the corresponding scale. Due to morphological filtering, the scale behavior of the regions is representable by a tree structure describing the spatial inter- and intra-scale relations among regions. The significance of a watershed region is automatically derived from its scale behavior by considering various attributes describing scale-dependent, morphological, and statistical properties of the region. The most significant regions from the segmentation of the image. The algorithm was verified for various medical image domains, such as cytological micrographs, bone x-rays, and cranial NMR slices.

Morphological multiscale shape analysis of light micrographs

Shape analysis of light-micrographs of cell populations is important for cytotoxicity evolution. This paper presents a morphological method for quantitative analysis of shape deformations of cells in contact to a biomaterial. After illumination normalization, a morphological multiscale segmentation yields separated cells. Shape deformation, and hence, toxicity of the substance under scrutiny, is quantified by means of compactness distribution and pattern spectrum of the population. Since the logarithmic image model is applicable to transmitted light, illumination normalization is achieved by removing the illumination component from the log- image by a tophat transform utilizing a large reconstruction filter. Subsequent thresholding and noise filtering yields connected binary cells, which are segmented by a marker-based, multiscale approach. For this, size-specific marker scales are generated removing noise and false markers. Each cell is now represented by an isolated marker. Converse integration of marker scales is performed by successive reconstruction of the original cell shapes, preventing merging of markers. Our method yields reasonable cell segmentations that go along with cell morphology even for differently sized and very distinct shapes. The obtained quantitative data is significantly correlated to the toxicity of the substance to be evaluated. Currently, the method is used for extensive biocompatibility tests.

Co-Occurrence Matrizen zur Texturklassifikation in Vektorbildern

Statistische Eigenschaften naturlicher Grauwerttexturen werden mit Co-Occurrence Matrizen, basierend auf der Grauwertstatistik zweiter Ordnung, modelliert. Die Matrix gibt dann die apriori Wahrscheinlichkeiten aller Grauwertpaare an. Da in der medizinischen Bildverarbeitung verstarkt Multispektralbilder ausgewertet werden, wird das bekannte Konzept hier auf beliebige Vektorbilder erweitert. Dadurch kann bei der Texturklassifikation die zur Verfugung stehende Information vollstandig genutzt werden. Insbesondere zur Detektion von Farbtexturen ist dieser Ansatz geeignet, da Wertepaare unterschiedlicher Spektralebenen ausgewertet werden konnen. Ebenso kann die Methode auch bei der Multiskalendekomposition von Intensitatsbildern zur Verbesserung der Texturerkennung beitragen. Die in den Matrizen entstehenden Muster lassen dann uber die Extraktion geeigneter Texturdeskriptoren Ruckschlusse auf die Texturen des Bildes zu.

Quantitative Messung der Hirnperfusion in intrakraniellen Ultraschall-Bildsequenzen

Die Darstellung und quantitative Auswertung der Hirnperfusion liefert wichtige diagnostische Hinweise. Mit der Computer-bzw. Magnet-Resonanz Tomographie stehen hierfur derzeit kosten- und zeitintensive Verfahren zur Verfugung. Demgegenuber stellt die Echo- densitometrie eine alternative, mobil einsetzbare Methode dar. Das dabei verwendete Ultraschallkontrastmittel besteht aus gasgefullten Mikroblaschen und bringt daher nur eine minimale Patientenbelastung mit sich. Verschiedene Faktoren schranken allerdings die Bildqualitat drastisch ein. Dieser Beitrag stellt eine neuartige Methode zur automatischen Quantifizierung der zerebralen Mikrozirkulation vor. Hierzu werden Harmonie Imaging Bildsequenzen mit stetig steigendem Aufnahmeintervall aufgenommen. Dadurch kann eine Sattigungskurve des Kontrastmittels im Blut ermittelt werden, aus deren Steigung sich die Flusgeschwindigkeit ergibt. Ein normiertes Mas fur die Hirnperfusion ist hieraus ableitbar. Weiterhin wird eine Verbesserung des Verfahrens vorgeschlagen, die auf den Musterspektrum der tiefpasgefilterten Bildsequenz basiert.

Formal extraction of biomedical objects by subgraph matching in attributed hierarchical region adjacency graphs

Extraction of objects from biomedical images is the fundamental task for many high level applications in medical image processing such as cytometry or diagnostic decision support. Therefore, a formal specification of sought objects is required along with an extraction procedure. On the basis of a hierarchical image decomposition objects are described by image regions of characteristic shape, texture, and visual context. For example, a cell consists of a circular core, a surrounding body containing organelles, which is in turn surrounded by the nutrition agent, and other cells. This is modeled by hierarchical graph representation of the region topology as nodes and the region properties as node attributes. In a hierarchical region representation, an object is described by subregions which again may contain subregions, thus object extraction becomes the matching of the respective region nodes. Obviously, graph matching is a NP-complete problem and therefore, it requires heuristics to become computable. This even holds for subtree matching. We propose a new approach which makes strongly use of the inclusion property of regions in a hierarchical image decomposition along with the visually descriptive attributes. The algorithm iterates a top-down bottom-up sequence over the region hierarchy to restrict the search space. Hence at each step, a layer of tree-node attributes must be compared to the attributes of the sought objects root node description. The bottom-up analysis is only invoked for the subtree depending on those nodes. Thus, each node is visited according to the topology of its visual occurrence in an image.