Local adaptive linear filters for image restoration were reported by the present author on one of the previous Wien workshops on image processing and computer graphics. The filters work in the domain of an orthogonal transform in a moving window and nonlinearly modify the transform coefficients to obtain an estimate of the central pixel of the window.In the present paper, we suggest an extension of local adaptive filtering in the domain of DFT/DCT to the processing multicomponent images, describe a recursive algorithm for local 3D DCT spectrum analysis which permits to substantially reduce the computational complexity of the filtering, present experimental result in denoising, enhancement and deblurring monochrome and color images and in target location in stereoscopic images. We also discuss using other than DCT orthogonal and color images and in target localization in stereoscopic images. We also discuss using other than DCT orthogonal transforms such as Walsh- Hadamard and Haar transforms and show that, for the window size 3 X 3 pixels, filtering in transform domain can be carried out in a form of 5 channel convolution with masks implementing local mean and 4 directional Laplacians: horizontal, vertical and two diagonal.

Gabor analysis is based on a well-structured composition of a signal or image as a series of building blocks. These are obtained by shifting and modulating a basic signal g along a time-frequency lattice (Lambda) . These families are typically non-orthogonal. Nevertheless suitable expansion coefficients of a signal f can be computed efficiently via the short time Fourier transform of f, using a canonically related window, the (Lambda) -dual atom g. A new iterative method to compute g is introduced, which is based on the Wexler-Raz principle and an appropriate version of POCS.

In this paper, we investigate a problem in computation of running discrete orthogonal transforms (RDOT). Due to overlapping blocks of signals, the RDOT for the signal block j can be computed recurrently by representing it as sum of tow terms. The first one is obtained by a multiplication of the circular advance matrix (CAM) by the RDOT of signal block j-1, and the second term is the transform of the sparse vector, formed from the weighted differences of the samples discarded from the block j-1 and the incoming samples of the block j. The second term of this RDOT decomposition could be efficiently implemented using fast prunned algorithms. The computational complexity of the RDOT depend mainly on the implementation of the first term. General conditions on the transform matrix for which CAM is completely sparse is established.

One of the earliest models of stochastic growth was originally developed for simulating the appearance of various biological patterns; in particular, bacterial colonies. Although it received little attention from biologists, some twenty years later it was adopted by crystallographers, solid state researchers, other physicists and chemists. Because of the model's flexibility it is being used by them after modifications appropriate to the application, in order to simulate their physical study objects under a variety of conditions. Only within the last few years has there been any interest in using this and similar digital models to represent the possible products of biological processes. It is also worth noting that aside from its relevance to probabilistically influenced pattern formation, the model has possible use in image processing for image compression and as an information-lossless way to code, regions, contours, or line segments.

The major goal of this survey is to provide the reader with the motivation of image filtering and segmentation in diagnostic imaging, with the brief overview of the state-of- the-art of nonlinear filters based on the geometry-driven diffusion (GDD), and with a possible generalization of the GDD-filtering towards the complex problem of image segmentation, stated as minimization of particular energy functionals. An example of the application of the GDD- filtering to the task of 3D visualization of MRI data of the brain is illustrated and discussed in the paper.

The presence of degraded or soft wedges on a scene or digital image may cause ambiguities in depth perception. This kind of edge has been analyzed by introducing a mathematical model to digitally implement a large category of degraded edges. This model has accounted the origin of degradation, the performance of psychophysical test for measuring the degradation perception, and image characteristics currently used in artificial vision as momenta, number of gray scale level per pixels, etc. It is the subject of current researches to be applied to artificial reproduction of human visual perception.

Two local adaptive nonlinear filters for image denoising are described and compared to other methods. The first one is a median filter computed over a connected neighborhood that fits the local conformation of level lines. The second one deals with the size of grains in image. Both filters depend on a single parameter, the area A, are morphological more than other classic morphological filters to denoise while preserving image structures. They are particularly designed for automation.

The major trends and developments in medical image processing are highlighted which are taking advantage of recent advances in computer science and technology. Through the wide-spread use of visual information processing within numerous fields of medicine a dramatical change of the health supply has taken place. The impact of the information processing technology on the treatment of patients are summarized.

The paper deals with geometry-driven diffusion filtering of MR tomograms of human brain. A method for making the conductance function locally adjustable has been proposed. It is based on using quantitative measures of pixel dissimilarities in neighborhoods which control the selection of an appropriate parameter of the exponential conductance. Three kinds of dissimilarity have been proposed and tested on a MR head phantom, as well as on real MR tomogram. A study of separate filtering effect on region interiors and boundaries has been performed using histograms of local conductance.

Visualization became an important branch of scientific research during the past decade. Some topics of visualization, for example, flow visualization and volume visualization, already provide lots of high-quality solutions to the most important problems. Others, as e.g., information visualization, are rather young disciplines which continuously provide lots of new advances. One such topic is the visualization over the Internet. Quite a number of approaches emerged during the last two years which facilitate this idea and embed the World Wide Web into the visualization process. Another trend in visualization is collaborative visualization within an augmented reality setup. Users exploit real 3D interaction and augmented reality to combine communication and investigation during research or education. Thirdly, the visualization of multi- dimensional and multi-modal data is a rather hot challenge to current scientific research. Coping with the restriction of just a few dimensions for representing visualization results is a big open problem in visualization and currently subject to eager research.

There is a wide range of visualization techniques for dynamical systems. These methods are used to visualize certain properties as, e.g., stability of fixed points, characteristic changes of velocity, and bifurcations. This paper gives a short introduction to dynamical system and describes several visualization techniques. Some of those are applied to three different dynamical system. The application of different visualization methods to dynamical systems shows how scientific visualization can be used for analyzing the behavior of dynamical systems, and how visualization can make analysis of a dynamical system fast and efficient.

This paper describes the design and implementation of VORTEX, a software package for generating high-quality images depicting shadowed and colored/textured representations of objects inherently present in a volumetric data set. It is realized as a C++-class library, provided with an API, and suited for inclusion in customized programs. Besides, VORTEX features a Tcl- interface allowing for interactive execution or preparation of scripts for calculating longer image sequences under controlled alteration of the viewing and/or rendering parameters. Specialties of the package include oversampling in image and object space, voxel traversal acceleration, dynamic adaptation of the ray/surface intersections, depth- dependent transparency, higher order gray level gradient approximations, arbitrarily orientable cut planes, and texture-mapped surface colorations. The design of the class library adheres strictly to the object-oriented paradigm which enables easy alteration and/or adaptation of additional classes and objects. The source and header files of the library are extensively documented according to an own scheme allowing for automatic generation of man-pages and Tcl-wrappers for the Tcl-interpreter by means of a suite of proprietary Perl-scripts. A practical application of VORTEX to a brain physiology problem is discussed at the end of the paper.

'Shadow profiling' measures shadow durations on an arbitrary scene during several hours of a specific day or even several weeks or months. The result is to be displayed visually. We shortly discuss already known techniques like simplified radiosity or discontinuity meshing with regard to their suitability for this problem. Due to various drawbacks of these techniques, we present our won approach. Especially a pixel-oriented version works very efficiently in connection with fast polygon-oriented shadow algorithms. It can be applied to architectural design, and it can also be used in computer graphics for the computation-inexpensive simulation of complex light sources.

Free-form deformation is an efficient, intuitive, and elegant method for solid geometric modeling and soft object animation. It is defined via a set of control points initially positioned at a regular user-defined 3D lattice, which embeds either the whole geometric model or just part of it. By successively deforming the 3D lattice, the model is deformed as well, e.g. into a bent, twisted or tapered model with a rather complex shape. Traditionally, the displacement of the control points is performed interactively until the final model meets the aesthetic requirements of the designer. On the other hand, e.g. in the field of practical engineering, the models have to fulfill physical requirements respectively quality criteria.In this case, the new positions of the control points need to be calculated, which is a very complex optimization task. We propose to use evolution strategies to solve this problem. The advantages and possibilities of this approach are demonstrated with one example, the calorimetric characterization of a scanner, where the results can directly be compared with the results; gained by a standard polynomial regression algorithm.

Construction of digital elevation models (DEMs) for natural earth surfaces have now become common practice. What does one do when the topography is caused by the closely-spaced tall buildings of lower Manhattan, New York City. Although the DEM principles are applicable, special problems were encountered in this very exciting and innovative project, requiring unique solutions, which are all discussed in this article.

An approach is proposed to the modeling and analysis of complex domains with a great diversity of object classes, where not all confusion situations can be previewed. The automatic recognition of these situations is shown. We also demonstrate the generation of consistent interpretations and their global evaluation.

The current starting satellite missions with a new generation of optical sensors will deliver very high resolution satellite imagery in the near future. Ground pixel sizes of 1 to 3 meters will give rise to new applications of remotely sensed imagery. Within forest stands, differentiation between single trees will be possible. Even parts of tree crowns will be recognizable. But, on the other hand, this will also give new challenge to satellite image analysis techniques. In this study, color IR aerial photos of forested land were used to simulate images of different ground pixel size. The effects of changing resolution on the spectral distribution properties of image regions representing different forest stands are investigated. Conclusions for automatic image segmentation and classification of forest stands are made.

Automatic pattern recognition by means of fuzzy logic had been applied to several fields during the last years. The spectral properties of different land cover types as seen in multiband images can also be interpreted as patterns in the dimension of gray values. Fuzzy clustering therefore is a new promising approach to mapping land cover from remotely send images. The traditional method of classifying a remotely sensed image is the transformation via a classification algorithm into a single classified image of the land surface, but natural landscapes present a continuum of variety at many different scales and a high proportion of the discretely sampled pixels within an image contains mixed spectral signature and are not easily placed into fixed thematic classes. The estimation of fuzzy memberships to vague classes of land cover more faithfully represents the true situation.

Satellite image segmentation with the aim to detect spatial units having an ecological meaning has become an important field in methodological research in modern landscape ecology. Within the theoretical framework, elaborated in, a landscape can be defined as spatial arrangement of ecosystems. Regions, that are more or less homogeneous in that sense, become more and more important as land units for physical-planning purposes. Such spatial objects can of course differ in size, in Central Europe they usually can be eliminated at the scale of some square kilometers. Therefore such objects should be detectable on satellite images and can then be ecologically characterized by their most important features: structure, function and change. Elaborating operational procedures to do that for the Austrian territory is the aim of a multidisciplinary research project called SINUS (structural features of landscapes as indicators for sustainable land use), which is financed by the Austrian Ministry for Science and Transportation and will be finished in 1999. This article can be regarded as one methodological output of this project.

An approach to the interpretation and recognition of images base on the directional field is developed. A model of an image with structure redundancy is introduced. Digital methods for the construction of the directional field, including the method of coal quadratic approximation, the spectral method, and the variance method, are considered. An optical scheme for the construction of the directional field with the use of radial segment spatial filters is presented. The developed approach is tested in the analysis of fingerprint images, tear crystallograms and interferograms.

The extraction of some unique for a given fingerprint reference point will considerably reduce the size of the description code and an indeterminate zone where an exhaustive matching of the minutiae sets must be performed, besides there is an opportunity not only to increase the accuracy of identification, but also to receive an exact criterion of availability ofpresented fingerprint in a database. The most essential in this case is the detection of reference point with accuracy not worse than up to 1/2 of an average period of structure (here the latter means an average shortest distance between distinct ridges). There are variety ofalgoritkms for singular points detection and automatic classification of fingerprints. The distinct features of our approach are the possibility to obtain very high accuracy, an opportunity to be applied before the stage of directional image generation, and ,in some cases, higher robustness to noises and distortions. Two main methods for singular reference point detection using Hough transform are proposed: clustering and relaxational, a classification of fingerprint based on a trajectory and dynamics of convergence being executed as a byproduct in the latter case. The offered algorithms were implemented in the test fingerprints identification system, the results obtained on the databases of 10000 and 500000 fingerprints have demonstrated accuracy of identification about 98% and 97% respectively. Keywords: image processing, directional image, fingerprint recognition, minutiae points.

One model for fingerprint's directional image simulation and two basic models for generating virtual ridges structure and minutiae are proposed. Both approaches are based on general physical principles. These models allow to explain natural processes of fingerprints appearance and offer some new opportunities for image regeneration in defected areas.

In this paper we represent a fingerprint by SPFL-graph with singular points, cores and deltas, as vertices: and fault lines as edges. We propose and prove some theories based on the SPFL-graph so that the types of singular point in fingerprints can be predicted. This is a totally new idea to fingerprints which will give a new direction of research in this field.

For feasible recognition having many categories such as Japanese character recognition, fast matching algorithms are necessary because the matching process occupies most of recognition time. In addition, for improving recognition accuracy, the matching process must use more complicated discrimination functions or a higher dimensional feature space, which involves higher computational costs. Therefore, pre-classification is used, which outputs a set of candidate categories to decrease the number of computations of the complicated discrimination functions.

A recurrent cellular neural network (CNN) for dot pattern clustering is presented. The CNN is based on dot pattern smoothing with a recursive algorithm which updates the dot positions by summing up nonlinear functions of position differences of Voronoi neighbors in each step. A perception criterion which assesses the proximity of dots by comparing their distance with an adaptive threshold is used for weighting the position difference sand for the generation of a special graph, the Dot Proximity Graph (DPG). The clusters are the connected components of the DPG. The method explains the Gestalt phenomenon and often corresponds with human perception which is demonstrated by representative examples. A possibility to deal with cluster hierarchies and dot textures is opened. Adequate cluster descriptions, especially a simple but powerful shape description, and a sequence of CNN's are needed for that purpose.

Burt introduced 1983 'equivalent weighting function': 'Iterative pyramid generation is equivalent to convolving the image g₀ with a set of 'equivalent weighting functions' h_l' g_l equals h_l * g₀ equals h * g_l-1, l > 1. It allowed him to study the effects of iterated reduction using the single parameter h_l without giving up the efficient iterative computation. A similar concept applies to graph pyramids built by dual graph contraction. This new algorithm reduces the number of vertices and of edges of a pair of dual image graphs while, at he same time, the topological relations among the 'surviving' components are preserved. Repeated application produces a stack of successively smaller graphs: a pari of dual irregular pyramids. The process is controlled by selected decimation parameters which consist of a subset of surviving vertices and associated contraction kernels. These pay a similar role for graph pyramids than the convolution kernels of Gaussian pyramids. Equivalent contraction kernels combine two or more contraction kernels int one single dual contraction. The basic concepts are elaborated and discussed. The new theory opens a large variety of possibilities to explore the domain of 'all' graph pyramids.

Optimal color quantization of true-color images is very important for various multimedia applications. We used MacAdam color space, where all color distances are Euclidean, to realize quantization girds that are optimal in respect to human vision. Simple fractionally-bilinear (FB) approximations are proposed for strictly non-linear MacAdam formulae that describe the transformation of usual color space to MacAdam space. Optimal coefficients of FB functions are found for the inner part of the color triangle. Using of FB functions gives the possibility to find explicit color quantization for true-color real-world images in MacAdam space. Both rectangular and hexagonal quantization grids were used in our experiments. Visual tests have shown that the quality of a true-color image displayed on the screen of computer monitor remained very high up to mesh sizes of 8-10 just noticeable differences. Color entropy of quantized images was in this case about 5-6. This means the possibility to compress their colors about 3 times with the help of standard statistical methods. More complicated compression methods that remove spatial correlation of the neighboring image pixels can be used to reach much higher compression ratios.

An image is assumed to be a blend of several statistically and semantically independent components, containing details of different information classes. Discussing about image decomposition we mean the splitting of an image onto the set of such components. Statistical distinctions of them allow to find effective algorithm for decomposition an image to several components of different properties. It gives the opportunity to extract form an image only the component of interest, thus to avoid redundant information from the following analysis, and finally to create decomposition- based image processing methods. To compare resulting images we introduce some formal quantitative measure for image estimation 2D variation, which is relied on continual image model. Its application for estimating of image decomposition is discussed. Under investigation we consider discrete model of image fragment, find rank algorithm for image decomposition, and discuss estimates of 2D variation.

The paper outlines our investigations devoted to creation of an efficient database of archive images of various types.Possible architecture of this database is briefly described. The attention is paid to methods and means of high-quality digital presentation of images, and to methods of their quality enhancement. The possibilities of compact and lossless storage of images are discussed. A prototype of the multifunctional database is described; this database can contain black-and-white, gray-scale, and true color images. It will be used to keep the representations of cultural treasures of various kinds, and to solve various problems of art-critics, linguistics, etc.

The paper presents results of research and development made to create an automated data base of archival photo- documents. The proposed approach is realized in a system consisting of three components: (1) a subsystem for document input; (2) a subsystem for digital image processing; and (3) an archiving subsystem. On the basis of the proposed approach a database prototype of archival photo-documents of Russian Academy of Sciences has been created. The work focuses on the development of economical and portable solutions suitable for both centralized and distributed database architecture allowing Web access.