Til Aach’s research while affiliated with University of Lübeck and other places

The detection of moving objects in image sequences acquired by a static camera is a demanding task in the presence of noise. With fast illumination variations detection becomes even more difficult since simple motion detection will always be subject to illumination artifacts. To solve these problems, we propose a Bayesian spatio-temporal motion detection algorithm combined with homomorphic filtering. The 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. Experimental results show that the presented method is considerably less sensitive to noise and time-varying scene illumination.

While oral lesions that macroscopically are highly suspicious for cancer shall be submitted to scalpel-biopsy and histologic evaluation, the majority of facultatively precancerous lesions, such as leuko- and erythroplakias or even persistent lichen planus lesions, may be assessed by brush-biopsy and cytology. As this non-invasive procedure is well tolerated by patients, more lesions may be screened and thus more oral cancers may be found in early, curable stages. Oral brush-biopsies can easily be performed by dentists, dental surgeons and general practitioners. While sensitivity of exfoliative cytology alone is about 4 % lower than bioptic histology, the combination of the latter with DNA-image-cytometry reaches the same diagnostic accuracy as the former (sensitivity 97.8–100 %, specificity, 97.4–100 %). As clonal chromosomal- and DNA-aneuploidy mostly precede cytological and histological evidence of malignancy in the squamous epithelium, its detection allows the diagnosis of oral squamous cell carcinomas up to 2 years earlier. Moreover, the additional use of DNA-image cytometry is a reasonable tool for the assessment of the resection margins of squamous cell carcinomas. DNA-image-cytometry could help to find the appropriate treatment option for patients and thus might improve their prognosis. Diagnostic DNA-image-cytometry is an objective method and has internationally been standardized. Here for the first time we describe the performance of a device for automated screening of oral brush-biopsy-smears, based on a cascaded diagnostic strategy: (1) Automated selection of morphologically suspicious nuclei, (2) Interactive control and correction of these on digital image galleries, (3) DNA-measurements on these only, (4) Final diagnosis on DNA-histograms and nuclear morphology.

The organization of the keratin intermediate filament cytoskeleton is closely linked to epithelial function. To study keratin network plasticity and its regulation at different levels, tools are needed to localize and measure local network dynamics. In this paper, we present image analysis methods designed to determine the speed and direction of keratin filament motion and to identify locations of keratin filament polymerization and depolymerization at subcellular resolution. Using these methods, we have analyzed time-lapse fluorescence recordings of fluorescent keratin 13 in human vulva carcinoma-derived A431 cells. The fluorescent keratins integrated into the endogenous keratin cytoskeleton, and thereby served as reliable markers of keratin dynamics. We found that increased times after seeding correlated with down-regulation of inward-directed keratin filament movement. Bulk flow analyses further revealed that keratin filament polymerization in the cell periphery and keratin depolymerization in the more central cytoplasm were both reduced. Treating these cells and other human keratinocyte-derived cells with EGF reversed all these processes within a few minutes, coinciding with increased keratin phosphorylation. These results highlight the value of the newly developed tools for identifying modulators of keratin filament network dynamics and characterizing their mode of action, which, in turn, contributes to understanding the close link between keratin filament network plasticity and epithelial physiology.

Das Narrow Band Imaging (NBI) erleichtert die Differenzierung zwischen Adenomen und harmlosen Darmwandgebilden (Hyperplasten) durch eine kontrastreiche Blutgefäßdarstellung. Dadurch wird eine automatische Segmentierung der Blutgefäße und Klassifizierung der Kolonpolypen hinsichtlich ihrer Entartungsfähigkeit ermöglicht. Eine erste Voraussetzung besteht in der automatisierten Trennung der Polypenoberfläche von der umliegenden Darmwand. In dieser Arbeit wird ein Verfahren vorgestellt, welches die Nachahmung der Fähigkeiten des Menschen zur Mustererkennung umsetzt. Die kombinierte Einbeziehung von Farbe und Textur gelingt mithilfe eines Gaußschen Farbmodells und Gaborfilterbasierten Aktiven Konturen. Eine Lokalisierungsgenauigkeit von 81 % und eine Spezifität von 94 % gewährleisten eine akkurate Polypensegmentierung durch die vorgestellte Level-Set-Methode und verprechen eine erfolgreiche Erkennung auch anderer Gewebe im Rahmen optischer sowie radiologischer Untersuchungen.

In this work we present a hierarchical segmentation algorithm for textured images, where the textures are composed of different number of additively superimposed oriented patterns. The number of superimposed patterns is inferred by evaluating orientation tensor based quantities which can be efficiently computed from tensor invariants such as determinant, minors and trace. Since direct thresholding of these quantities leads to non-robust segmentation results, we propose a graph cut based segmentation approach. Our level dependent energy functions consist of a data term evaluating orientation tensor based quantities, and a smoothness term which assesses smoothness of the segmentation results. We present the robustness of the approach using both synthetic and real image data.

The linear periodically shift variant (LPSV) properties within critically sampled multirate FIR filter banks are generally analysed using deterministic signals. Periodic shift variance is, however, closely related to cyclostationarities introduced by the LPSV system into originally wide sense stationary (WSS) random signals passing through the system. We provide first a unified framework to measure both shift variance of the LPSV system and the amount of cyclostationarity it generates. In this respect, the key concept is the covariance operator associated to a random variable. Cyclostationarity of the variable translates to LPSV properties of the operators, and vice versa. We study several related concepts for the quantification of shift variance in operators and their interpretation in the stochastic setting. We then introduce a new concept called expected shift variance. Fourier-analytic expressions for the various measures are given, and subsequently used to derive explicit formulae for the case of critically sampled two-channel filter banks, as well as sharp upper bounds for unitary two-channel filter banks. Numerical evaluations of the measures show that they provide largely consistent rankings between various critically sampled two-channel filter banks.

Fluorescence confocal laser scanning microscopy (CLSM) has revolutionized imaging of subcellular structures in biomedical research by enabling the acquisition of 3D time-series of fluorescently-tagged proteins in living cells, hence forming the basis for an automated quantification of their morphological and dynamic characteristics. Due to the inherently weak fluorescence, CLSM images exhibit a low SNR. We present a novel model for the transfer of signal and noise in CLSM that is both theoretically sound as well as corroborated by a rigorous analysis of the pixel intensity statistics via measurement of the 3D noise power spectra, signal-dependence and distribution. Our model provides a better fit to the data than previously proposed models. Further, it forms the basis for (i) the simulation of the CLSM imaging process indispensable for the quantitative evaluation of CLSM image analysis algorithms, (ii) the application of Poisson denoising algorithms and (iii) the reconstruction of the fluorescence signal.

Registration of an image non-rigidly to another one causes deformations, which generally do not preserve the initial volume. Volume preservation is however indispensable for observing tumors in medical images. This paper presents the correction of B-spline based registration to preserve the volume in observed regions. In contrast to other approaches, our solution is not obtained through energy minimization, but by calculating the correction parameters for the deformation directly. Especially for high resolution image data this strategy is very efficient in terms of computational expenses. We derive a closed form solution to optimize the registration with respect to the compression at a single point and then extend the problem to multiple points. Finally we prove also that the correction terms do not have any significant influence on the registration quality.

A prototype system for automatic in-line flaw detection in industrial woven fabrics is presented. Where state of the art systems operate on low-resolved (≈ 200 ppi) image data, we describe here the process flow to segment single yarns in high-resolved (≈ 1000 ppi) textile images. This work is partitioned into two parts: First, mechanics, machine integration, vibration cancelling and illumination scenarios are discussed based on the integration into a real loom. Subsequently, the software framework for high precision fabric defect detection is presented. The system is evaluated on a database of 54 industrial fabric images, achieving a detection rate of 100% with minimal false alarm rate and very high defect segmentation quality.

Machines for automated PCB depaneling have greatly improved the industrial production efficiency of electronic products. But the preparation for automated depaneling could be very complex and time-consuming. In this paper, we propose a novel systematic solution for this problem. Using a vision-based assistant system, called ”Auto-Teaching”, all connection tabs that should be milled off panels are detected automatically. Highly accurate milling curves are generated with respect to the geometry of the corresponding tabs. Then they could be easily converted into CNC code which drive milling cutters later. Moreover, placement suggestions of supporting pins used to fix panels are obtained. All image analysis functions are implemented in C++ and optimized to meet the minimal hardware and time requirements. Thus the whole process of the preparation is simplified for users and the PCB manufactures can benefit from reduced idle running of machines and labor costs.