Horst Haußecker

Computer vision and applications: a guide for students and practitioners

Preface. Contributors. B. Jahne, Introduction. Sensors and Imaging. H. Haubecker,Radiation and Illumination. P. Geibler,Imaging Optics. H. Haubecker,Radiometry of Imaging. P. Seitz,Solid-State Image Sensing. R. Godding, Geometric Calibration of Digital Imaging Systems. R. Schwarte, G. Hausler, R.W. Malz, Three-Dimensional Imaging Techniques. Signal Processing and Pattern Recognition. B. Jahne, Representation of Multidimensional Signals. B. Jahne, Neighborhood Operators. H. Haubecker, H. Spies, Motion. P. Geibler, T. Dierig, H.A. Mallot, Three-Dimensional Imaging Algorithms. J. Weickert, Design of Nonlinear Diffusion Filters. C. Schnorr, Variational Adaptive Smoothing and Segmentation. P. Soille, Morphological Operators. J. Hornegger, D. Paulus, H. Niemann,Probabilistic Modeling in Computer Vision. H. Haubecker, H.R. Tizhoosh, Fuzzy Image Processing. A. Meyer-Base, Neural Net Computing for Image Processing. Application Gallery. Index.

Fuzzy image processing

Publisher Summary This chapter provides an overview of the basic principles and potentials of state of the art fuzzy image processing that can be applied to a variety of computer vision tasks. The world is fuzzy, and so are images, projections of the real world onto the image sensor. Fuzziness quantifies vagueness and ambiguity, as opposed to crisp memberships. The types of uncertainty in images are manifold, ranging over the entire chain of processing levels, from pixel based grayness ambiguity over fuzziness in geometrical description up to uncertain knowledge in the highest processing level. The human visual system has been perfectly adapted to handle uncertain information in both data and knowledge. The interrelation of a few such “fuzzy” properties sufficiently characterizes the object of interest. Fuzzy image processing is an attempt to translate this ability of human reasoning into computer vision problems as it provides an intuitive tool for inference from imperfect data. Fuzzy image processing is special in terms of its relation to other computer vision techniques. It is not a solution for a special task, but rather describes a new class of image processing techniques. It provides a new methodology, augmenting classical logic, a component of any computer vision tool. A new type of image understanding and treatment has to be developed. Fuzzy image processing can be a single image processing routine or complement parts of a complex image processing chain.

A Tensor Approach for Precise Computation of Dense Displacement Vector Fields

Using the 3-dimensional structure tensor, dense displacement vector fields (DVF) can be computed with subpixel accuracy. The approach is based on the detection of linear symmetries and the corresponding orientation angle within a local spatio-temporal neighborhood. The proposed technique is well suited to be applied within a multiresolution framework in order to avoid corresponence problem errors. In this paper the implementation of the tensor approach will be introduced together with an error analysis. The performance of the technique will be proved with test sequences as well as with real scientific applications.

Differential Range Flow Estimation

We present a total least squares based differential method for the estimation of 3D range flow from a sequence of range images. We address the various manifestations of the aperture problem encountered with this type of data. It is described how they can be detected and how the appropriate normal flow can be computed. The performance of the proposed method is assessed on both synthetic and real data.

Performance Characteristics of Low-level Motion Estimators in Spatiotemporal Images

This chapter presents an analytical, numerical, and experimental study of the performance of low-level motion estimators in spatiotemporal images. Motivation for this work arose from scientific applications of image sequence processing within the frame of an interdisciplinary research unit. Here, the study of transport, exchange, and growth processes with various imaging techniques requires highly accurate velocity estimates (Jahne et al., 1996; Jahne et al., 1998). These high accuracy demands triggered a revisit of the fundamentals of motion estimation in spatiotemporal images. In this chapter only low-level motion estimators are discussed. This is only a part of the picture, but errors in low-level estimators propagate and thus cause also errors in higher-level features. Only a few systematic studies of the performance characteristics of low-level motion estimators are available in the literature. Kearney et al. (1987) performed an error analysis of optical flow estimation with gradient-based methods, while Simoncelli (1999) studied the error propagation of multi-scale differential optical flow. Barron et al. (1994) used a set of computer-generated and real image sequence to compare various approaches to optical flow computation. To study the performance of phase-based and energy-based techniques, Haglund and Fleet (1994) used an image sequence generated by warping a single natural image. Otte and Nagel (1994) were the first to verify motion estimators with a calibrated real-world sequence. Bainbridge-Smith and Lane (1997) theoretically compared various first and second-order differential techniques and proved the results using a series of test sequences.

Radiation and Illumination

Publisher Summary This chapter discusses that visual perception of scenes depends on appropriate illumination to visualize objects. The human visual system is limited to a very narrow portion of the spectrum of electromagnetic radiation, called light. In some cases natural sources, such as solar radiation, moonlight, lightning flashes, or bioluminescence, provide sufficient ambient light to navigate our environment. Because humankind was mainly restricted to daylight, one of the first attempts was to invent an artificial light source—fire. Computer vision is not dependent upon visual radiation, fire, or glowing objects to illuminate scenes. As soon as imaging detector systems became available, other types of radiation were used to probe scenes and objects of interest. Recent developments in imaging sensors cover almost the whole electromagnetic spectrum from x-rays to radiowaves. In standard computer vision applications, illumination is frequently taken as given and optimized to illuminate objects evenly with high contrast. Such setups are appropriate for object identification and geometric measurements. Radiation, however, can also be used to visualize quantitatively physical properties of objects by analyzing their interaction with radiation. The chapter provides the fundamentals for a quantitative description of radiation emitted from sources, as well as the interaction of radiation with objects and matter. It also shows, using a few selected examples, how this knowledge can be used to design illumination setups for practical applications such that different physical properties of objects are visualized.

Rekonstruktion von Schreiberkurven

In der vorliegenden Arbeit wird eine Methode zur Rekonstruktion von Daten einer Schreiberkurve vorgestellt. Die Daten liegen in Form von Bildern der Schreiberkurve vor. Die Rekonstruktion der Daten wird mit Hilfe einer Segmentierung der Bilder im Farbraum vorgenommen, wobei der verwendete Algorithmus unabhangig von der Grose der Bilder ist. Die Bilder konnen mit hoher Genauigkeit digitalisiert werden. Zusatzlich sind auch Korrekturmoglichkeiten vorgesehen, um das Ergebnis zu uberprufen und ggf. zu korrigieren.