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.

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.

Image analysis problems such as feature tracking, edge detection, image enhancement, or texture analysis require thedetection of multi-oriented patterns which can appear at arbitrary orientations. Direct rotated matched filtering for feature detection is computationally expensive, but can be sped up with steerable filters. So far, steerable filter approaches were limited to only one direction. Many important low-level image features are, however, characterized by more than a single orientation. We therefore present here a framework for efficiently detecting specific multi-oriented patterns with arbitrary orientations in grayscale images. The core idea is to construct multisteerable filters by appropriate combinations of single-steerable filters. We exploit that steerable filters are closed under addition and multiplication. This allows to derive a design guide for multisteerable filters by means of multivariate polynomials. Furthermore, we describe an efficient implementation scheme and discuss the use of weighting functions to reduce angular oscillations. Applications in camera calibration, junction analysis of images from plant roots, and the discrimination of L, T, and X-junctions demonstrate the potential of this approach.

Endoscopic panorama images provide a wide field of view and assist in documentation and navigation. Their performance, registration errors, and blending artifacts have been evaluated. However, the similarity between local panorama regions and their input images based on a quality assessment regarding the human visual system (HVS) has not been addressed. Particularly, panorama regions composed by freehand sequences with varying image quality can result in less contrast and illumination. Thus, we develop a new fidelity score for quantitative image quality assessment based on structural similarity maps adopted to the HVS. The measure indicates how much structural information of relevant structures is preserved in the panorama. For evaluation, panoramas are generated from cystoscopic fluorescence videos, and three blending algorithms are evaluated. Our results show that the fidelity scores are consistent with visual assessments, and can be used to assess blending methods for panorama images specific to the given video sequence.

Examining the growth rate of pleural thickenings in consecutive 3D-CT images requires the matching of identical thickenings in lung images acquired at two different points in time. The thickenings can be subject to strong deformations caused by their growth. This implies that position information should play a major role in finding correspondences. Here, a MGRF approach is presented to determine a rigid transformation. It aligns the lung volumes by maximizing the probability of the regarded lung tissue to fit an offline trained model. To ensure a symmetrical matching of lung surfaces this probability is calculated reciprocally. Using precalculation, strong sub-sampling and a multiscale approach, the required time can be reduced by a factor of about 80, depending on the image resolution. Due to this speed-up, online follow-up assessment is feasible. We show that this approach results in precise registrations which can be used for a reliable matching of lung thickenings.

High-fidelity color image acquisition with a multispectral camera utilizes optical filters to separate the visible electromagnetic spectrum into several passbands. This is often realized with a computer-controlled filter wheel, where each position is equipped with an optical bandpass filter. For each filter wheel position, a grayscale image is acquired and the passbands are finally combined to a multispectral image. However, the different optical properties and non-coplanar alignment of the filters cause image aberrations since the optical path is slightly different for each filter wheel position. As in a normal camera system, the lens causes additional wavelength-dependent image distortions called chromatic aberrations. When transforming the multispectral image with these aberrations into an RGB image, color fringes appear, and the image exhibits a pincushion or barrel distortion. In this paper, we address both the distortions caused by the lens and by the filters. Based on a physical model of the bandpass filters, we show that the aberrations caused by the filters can be modeled by displaced image planes. The lens distortions are modeled by an extended pinhole camera model, which results in a remaining mean calibration error of only 0.07 pixels. Using an absolute calibration target, we then geometrically calibrate each passband and compensate for both lens and filter distortions simultaneously. We show that both types of aberrations can be compensated and present detailed results on the remaining calibration errors.

Medical investigations targeting a quantitative analysis of the position emission tomography (PET) images require the incorporation of additional knowledge about the photon attenuation distribution in the patient. Today, energy range adapted attenuation maps derived from computer tomography (CT) scans are used to effectively compensate for image quality degrading effects, such as attenuation and scatter. Replacing CT by magnetic resonance (MR) is considered as the next evolutionary step in the field of hybrid imaging systems. However, unlike CT, MR does not measure the photon attenuation and thus does not provide an easy access to this valuable information. Hence, many research groups currently investigate different technologies for MR-based attenuation correction (MR-AC). Typically, these approaches are based on techniques such as special acquisition sequences (alone or in combination with subsequent image processing), anatomical atlas registration, or pattern recognition techniques using a data base of MR and corresponding CT images. We propose a generic iterative reconstruction approach to simultaneously estimate the local tracer concentration and the attenuation distribution using the segmented MR image as anatomical reference. Instead of applying predefined attenuation values to specific anatomical regions or tissue types, the gamma attenuation at 511 keV is determined from the PET emission data. In particular, our approach uses a maximum-likelihood estimation for the activity and a gradient-ascent based algorithm for the attenuation distribution. The adverse effects of scattered and accidental gamma coincidences on the quantitative accuracy of PET, as well as artifacts caused by the inherent crosstalk between activity and attenuation estimation are efficiently reduced using enhanced decay event localization provided by time-of-flight PET, accurate correction for accidental coincidences, and a reduced number of unknown attenuation coefficients. First results achieved with measured whole body PET data and reference segmentation from CT showed an absolute mean difference of 0.005 cm⁻¹ (< 20%) in the lungs, 0.0009 cm⁻¹ (< 2%) in case of fat, and 0.0015 cm⁻¹ (< 2%) for muscles and blood. The proposed method indicates a robust and reliable alternative to other MR-AC approaches targeting patient specific quantitative analysis in time-of-flight PET/MR.

The B-spline mathematical functions have long been utilized for signal representation, zooming and interpolation. However they have not been investigated for different image coding applications. In spite of the fact that B-splines are a semi-orthogonal basis (not full orthogonal) they still can be utilized for redundancy removal and decorrelation maximization prior to any compression scheme. This is due to their distinctive feature of preserving energy in low frequencies. In this paper we propose a novel technique for preprocessing signals/images prior to the decomposition stage in different image coders based on the B-spline decomposition. Mathematical explanation and derivation for the proposed B-spline decomposition basis is presented and analyzed. We derive our theoretic/mathematical justification, through some Eigen analysis calculations, for the enhancement in compression performance achieved with our B-spline based compression approach. We also present a straightforward approach for calculating the B-spline basis in a fast and efficient manner. Extensive simulations have been carried out on the well known SPIHT image coder with and without the proposed correlation removal methodology. Simulation results that demonstrate the effectiveness of the proposed technique are presented.

Recently, visual representations using high dynamic range (HDR) images become increasingly popular, with advancement of technologies for increasing the dynamic range of image. HDR image is expected to be used in wide-ranging applications such as digital cinema, digital photography and next generation broadcast, because of its high quality and its powerful expression ability. HDR imaging technologies will spread its sphere of influence in imaging industry. In this paper, we review the state-of-the-art studies and the trends of the HDR imaging, in terms of the following three points: (1) HDR imaging sensor and HDR image generation techniques as image acquisition technologies, (2) encode method of HDR images for efficient transmission and storage, (3) human visual system issues associated with reproduction of HDR image.

High dynamic range (HDR) imaging techniques have been applied more and more in recent years. They are mostly based on the basic principle of acquiring differently exposed images of the scene, linearizing each image, and finally combining the resulting exposure set into one HDR image using some weighting scheme. This process is affected by different noise sources during image acquisition. In this paper we comprehensively study the noise reduction properties of HDR imaging. We show that the fact that individual images partially exhibit the same structures with independent noise can be utilized to further improve the signal-to-noise ratio (SNR). To this end, we model the whole imaging process and evaluate resulting HDR images using artificially created as well as real-world scenes. We compare imaging integrated denoising to denoising of the LDR input data as well as denoising the resulting HDR images directly in terms of SNR gain and detail loss.

We extend estimation of range flow to handle brightness changes in image data caused by inhomogeneous illumination. Standard range flow computes 3D velocity fields using both range and intensity image sequences. Toward this end, range flow estimation combines a depth change model with a brightness constancy model. However, local brightness is generally not preserved when object surfaces rotate relative to the camera or the light sources, or when surfaces move in inhomogeneous illumination. We describe and investigate different approaches to handle such brightness changes. A straightforward approach is to prefilter the intensity data such that brightness changes are suppressed, for instance, by a highpass or a homomorphic filter. Such prefiltering may, though, reduce the signal-to-noise ratio. An alternative novel approach is to replace the brightness constancy model by 1) a gradient constancy model, or 2) by a combination of gradient and brightness constancy constraints used earlier successfully for optical flow, or 3) by a physics-based brightness change model. In performance tests, the standard version and the novel versions of range flow estimation are investigated using prefiltered or nonprefiltered synthetic data with available ground truth. Furthermore, the influences of additive Gaussian noise and simulated shot noise are investigated. Finally, we compare all range flow estimators on real data.

Texture classification generally requires the analysis of patterns in local pixel neighborhoods. Statistically, the underlying processes are comprehensively described by their joint probability density functions (jPDFs). Even for small neighborhoods, however, stable estimation of jPDFs by joint histograms (jHSTs) is often infeasible, since the number of entries in the jHST exceeds by far the number of pixels in a typical texture region. Moreover, evaluation of distance functions between jHSTs is often computationally prohibitive. Practically, the number of entries in a jHST is therefore reduced by considering only two-pixel patterns, leading to 2D-jHSTs known as cooccurrence matrices, or by quantization of the gray levels in local patterns to only two gray levels, yielding local binary patterns (LBPs). Both approaches result in a loss of information. We introduce here a framework for supervised texture classification which reduces or avoids this information loss. Local texture neighborhoods are first filtered by a filter bank. Without further quantization, the jPDF of the filter responses is then described parametrically by gaussian mixture models (GMMs). We show that the parameters of the GMMs can be reliably estimated from small image regions. Moreover, distances between the thus modelled jPDFs of different texture patterns can be computed efficiently in closed form from their model parameters. We furthermore extend this texture descriptor to achieve full invariance to rotation. We evaluate the framework for different filter banks on the Brodatz texture set. We first show that combining the LBP difference filters with the GMM-based density estimator outperforms the classical LBP approach and its codebook extensions. When replacing these-rather elementary-difference filters by the wavelet frame transform (WFT), the performance of the framework on all 111 Brodatz textures exceeds the one obtained more recently by spin image and RIFT descriptors by Lazebnik et al.

High dynamic range (HDR) imaging is used to acquire the full dynamic range of a scene with a camera of limited dynamic range. To this end, an exposure set of the scene is acquired, followed by the linearization of each image with the inverse camera transfer function (CTF), which needs to be measured or estimated. Subsequently, the images are combined into one HDR image. Several weighting functions have been proposed for this combination. Naturally, each individual image is afflicted with noise from the acquisition process. The resulting HDR image features a higher SNR than the acquired images as a consequence of the weighted averaging during reconstruction. We show that the fact that the individual images partially show the same structures with independent noise can be utilized to further improve the SNR. Thus, we propose a wavelet-based denoising using correlation analysis between different images from the exposure set that outperforms the denoising properties of commonly applied weighted averages.

Cell adhesion, a process mediated by the formation of discrete structures known as focal adhesions (FAs), is pivotal to many biological events including cell motility. Much is known about the molecular composition of FAs, although our knowledge of the spatio-temporal recruitment and the amount of the individual components present in the FAs is still incomplete. To fill this gap, an essential prerequisite is a reliable procedure for the recognition, segmentation, and tracking of FAs. While manual segmentation and tracking may provide some advantages when done by an expert, its performance is usually hindered by subjective judgement and long time required to analyse large data sets. Here, we develop a model-based segmentation and tracking algorithm that overcomes these problems. We found that our algorithm matches FA segmentation done by an expert. In addition, it contains a dedicated computational approach to correct single frame segmentation errors that may arise from the analysis of poorly defined FAs. We believe that our work establishes the basis for a comprehensive qualitative and quantitative analysis of FA dynamics and the later development of mathematical models that simulate FAs behaviour.

Estimation of local orientations in multivariate signals is an important problem in image processing and computer vision. This general problem formulation also covers optical flow estimation, which can be regarded as orientation estimation in space-time-volumes. Modelling a signal using only a single orientation, however, is often too restrictive, since occlusions and transparencies occur frequently, thus necessitating the modelling and analysis of multiple orientations. We, therefore, develop a unifying mathematical model for multiple orientations: Beyond describing an arbitrary number of orientations in scalar- and vector-valued image data such as color image sequences, it allows the unified treatment of additively and occludingly superimposed oriented structures as well as of combinations of these. Based on this model, we describe estimation schemes for an arbitrary number of additively or occludingly superimposed orientations in images. We confirm the performance of our framework on both synthetic and real image data.

In flat detector cone-beam computed tomography and related applications, sparse angular sampling frequently leads to characteristic streak artifacts. To overcome this problem, it has been suggested to generate additional views by means of interpolation. The practicality of this approach is investigated in combination with a dedicated method for angular interpolation of 3-D sinogram data. For this purpose, a novel dedicated shape-driven directional interpolation algorithm based on a structure tensor approach is developed. Quantitative evaluation shows that this method clearly outperforms conventional scene-based interpolation schemes. Furthermore, the image quality trade-offs associated with the use of interpolated intermediate views are systematically evaluated for simulated and clinical cone-beam computed tomography data sets of the human head. It is found that utilization of directionally interpolated views significantly reduces streak artifacts and noise, at the expense of small introduced image blur.

High dynamic range (HDR) imaging is more and more widely used to increase the limited dynamic range of digital cameras and, in turn, to cover the dynamic range of the acquired scene. This image acquisition process can be subdivided into two steps. The first step is the measurement or estimation of the mostly non-linear camera transfer function (CTF). This is followed by the second step, the combination of a set of differently exposed images of the same scene into one HDR image after linearization with the inverse CTF. Each of the individual images in such an exposure set contains noise from the image acquisition process. Consequently, the calculated HDR image will as well contain noise, which fortunately is reduced by the weighted average of the images from the exposure set. We analyze the achieved gain in SNR for different weighting functions proposed in the literature and compare these with a plain average. Although these functions are based on reasonable intuitions, we show that the highest SNRgainis achieved with the plain average.

Multispectral cameras enable high-fidelity color image acquisition by separation of the electromagnetic spectrum with optical bandpass filters mounted on a filter wheel between optics and imaging sensor. A multispectral image is acquired by capturing one grayscale image per wheel position and combining the images afterwards. Since the optical filters differ in their thicknesses and refraction indices, both transversal (geometric distortions) and longitudinal aberrations (blurring) occur. We present a physical model for the longitudinal aberrations and describe the resulting distortion effects. This knowledge serves as a basis for our compensation algorithm, which performs a shift-variant deconvolution of the image, significantly improving the image quality. We furthermore describe a practicable calibration procedure for our algorithm.

Bladder tumors are difficult to recognize during an endoscopy of the bladder (cystoscopy) using a conventional white light source. Therefore, a technique called photo dynamic diagnosis (PDD) or fluorescence endoscopy is used to enhance both sensitivity and specificity of tumor detection. As preparation for a PDD cystoscopy, a marker substance is brought into the patient's bladder which leads to a fluorescence of cancerous tissue if a suitable light source is used. Even though tumors are better distinguishable from healthy bladder tissue, the contrast between both types of tissue can still be low and it further decreases with the decay of the marker substance. To overcome these difficulties, we enhance the contrast between fluorescence and background. To this end, we perform a color space transformation which separates fluorescence information from the remaining information. In this new color space, we apply a non-linear transformation to the fluorescence component which leads to a better separation of fluorescing and non-fluorescing pixels. Finally, the image is transferred back to the RGB color space. For clinical evaluation, we implemented the proposed algorithm as plug-in for the real-time capable software frameworkRealTimeFrame.

Multispectral image acquisition considerably improves color accuracy in comparison to RGB technology. A common multispectral camera design concept features a filter-wheel consisting of six or more optical bandpass filters. By shifting the filters sequentially into the optical path, the electromagnetic spectrum is acquired through the channels, thus making an approximate reconstruction of the spectrum feasible. However, since the optical filters exhibit different thicknesses, refraction indices and may not be aligned in a perfectly coplanar manner, geometric distortions occur in each spectral channel: The reconstructed RGB images thus show rainbow-like color fringes. To compensate for these, we analyze the optical path and derive a mathematical model of the distortions. Based on this model we present two different algorithms for compensation and show that the color fringes vanish completely after application of our algorithms. We also evaluate our compensation algorithms in terms of accuracy and execution time.

For successful cure, cancer has to be detected as early as possible. Since cancer starts from a single cell, this can best be done using cytopathological methods. One important diagnostically relevant measure is the proliferation rate of the cells, which can be estimated from segmented silver stained nuclei. However, the microscopy images of silver stained specimens vary strongly in intensity and contrast and are furthermore compromised by an overall texture. We show that a precise segmentation of the nuclei is possible using a two-step approach. First, an oversegmentation with the mean shift algorithm is obtained. In a second step, these regions are merged to objects, guided by a suitable shape model, viz an ellipse, but simultaneously allowing deviations from this shape model. The segmentation results are compared to a gold standard of 8617 nuclei from 23 specimens of the thyroid gland, achieving a mean areal segmentation error of DeltaA macrnucleus= 12mum2per nucleus.

Flat panel X-ray detectors allow the immediate availability of the acquired images for display. However, they provide images with defective areas amongst others due to manufacturing problems. In this contribution, we present a frequency selective extrapolation method in order to restore these defects by extending the surrounding signal into the defective area. In case of static radiographs, the spatial surrounding is evaluated by a 2-D approach. Defects in sequences acquired by cine-angiography or fluoroscopy are processed by 3-D extrapolation. The defects are replaced by extrapolating the signal from the spatial and at the same time temporal surrounding, taking previous and subsequent frames into account. Hence, inherent motion compensation is accomplished. The application of the spatial and where applicable spatio-temporal extrapolation approach allows to restore smooth areas, edges, patterns as well as noise. The ability to restore noise is especially important for medical images because it leads to a natural appearance of the concealed defective areas.

The theory of steerable filters (introduced by Freeman and Adelson in 1991) offers a convenient and computationally efficient method to represent the rotation of a specific class of 2D filters as a linear combination of base filters. In this paper, we generalize this theory to filters which are modeled as point-by-point products of multiple (in our examples: two) base filters which each are rotated by individual rotation angles, thus introducing the term "multi-steerability". While standard steerable filters are useful for the detection of edges or lines (oriented in some unknown direction), our novel extended theory can be applied to the detection of features which are characterized by multiple orientations, for instance corners or line crossings, which greatly increases the applicability of the theory. We conclude our paper with an example for detecting the crossings in a checkerboard image using double-steerable filters.

Silver staining of cytopathologic specimens offers advantages in cancer diagnostics. A difficulty with such stained cell specimens is the very high dynamic range needed by the imaging system to appropriately cover the varying stain intensities. Beside those images of cell nuclei that can be used for the diagnostic interpretation, there are nuclei that appear too dark to observe their relevant properties, the so-called argyrophilic nucleolar organizer regions (AgNORs), which appear as spot-like areas darker than their immediate surroundings. We therefore show how high dynamic range images of nuclei can help to correctly segment the AgNORs. To this end, we acquire a sequence of differently exposed images, which are then combined into a high dynamic range image. Based on the dynamic range of the image signal within the segmented cell area, we compute another image which provides optimal contrast over this area of interest. To further increase the contrast for dark objects, a suitable nonlinear point transform is simultaneously applied. We provide examples of the thus generated images and their corresponding segmentations.

Estimation of local orientation in images may be posed as the problem of finding the minimum gray-level variance axis in a local neighborhood. In bivariate images, the solution is given by the eigenvector corresponding to the smaller eigenvalue of a 2 x 2 tensor. For an ideal single orientation, the tensor is rank-deficient, i.e., the smaller eigenvalue vanishes. A large minimal eigenvalue signals the presence of more than one local orientation, what may be caused by non-opaque additive or opaque occluding objects, crossings, bifurcations, or corners. We describe a framework for estimating such superimposed orientations. Our analysis is based on the eigensystem analysis of suitably extended tensors for both additive and occluding superpositions. Unlike in the single-orientation case, the eigensystem analysis does not directly yield the orientations, rather, it provides so-called mixed-orientation parameters (MOPs). We, therefore, show how to decompose the MOPs into the individual orientations. We also show how to use tensor invariants to increase efficiency, and derive a new feature for describing local neighborhoods which is invariant to rigid transformations. Applications are, e.g., in texture analysis, directional filtering and interpolation, feature extraction for corners and crossings, tracking, and signal separation.

The success of skin transplantations depends on the adequate revascularization of the transplanted dermal matrix. To induce vessel growth or angiogenesis, pharmacological substances may be applied to the dermal matrix. The effectiveness of different such substances has been evaluated in laboratory experiments. For this purpose, the surface and length of newly grown vessels have to be measured in micro-angiograms (x-ray images of the blood vessels recorded after the injection of a radiopaque substance) of tissue transplanted on the back of laboratory animals. To this end we describe in this contribution a vessel analysis environment central to which is a vessel segmentation tool for surface quantification in fasciocutaneous skin transplant micro-angiograms.

The success of skin transplantations depends on a proper revascularization of the transplanted tissue. Angiogenesis (vessel growth) in the transplanted dermal matrices can be stimulated by administering different drugs. To evaluate the effectiveness of different drug treatments in an experimental setting using laboratory animals, vessels in transplant samples are visualized by so-called micro-angiograms, i.e., X-ray images after contrast agent injection. We describe a framework for the acquisition of such micro-angiograms as well as for the subsequent semi-automatic analysis of angiogenesis. Central to our analysis is the segmentation of small fasciocutaneous vessels in the transplant sample.

Local orientation estimation can be posed as the problem of finding the minimum grey level variance axis within a local neighbourhood. In 2D image signals, this corresponds to the eigensystem analysis of a 2 /spl times/ 2-tensor, which yields valid results for single orientations. We describe extensions to multiple overlaid orientations, which may be caused by transparent objects, crossings, bifurcations, corners etc. Multiple orientation detection is based on the eigensystem analysis of an appropriately extended tensor, yielding so-called mixed orientation parameters. These mixed orientation parameters can be regarded as another tensor built from the sought individual orientation parameters. We show how the mixed orientation tensor can be decomposed into the individual orientations by finding the roots of a polynomial. Applications are, e.g., in directional filtering and interpolation, feature extraction for corners or crossings, and signal separation.

This paper deals with the problem of estimating multiple motions at points where these motions are overlaid. We present a new approach that is based on block-matching and can deal with both transparent motions and occlusions. We derive a block-matching constraint for an arbitrary number of moving layers. We use this constraint to design a hierarchical algorithm that can distinguish between the occurrence of single, transparent, and occluded motions and can thus select the appropriate local motion model. The algorithm adapts to the amount of noise in the image sequence by use of a statistical confidence test. The algorithm is further extended to deal with very noisy images by using a regularization based on Markov Random Fields. Performance is demonstrated on image sequences synthesized from natural textures with high levels of additive dynamic noise.

We extend a novel framework for the estimation of multiple transparent motions to include regularization. We use mixed-motion parameters to obtain linear Euler-Lagrange equations with a regularization term. The equations are solved iteratively for the mixed-motion parameters based on an update rule that is similar to the case of only one motion. The motion parameters are then obtained as the roots of a complex polynomial of a degree that is equal to the number of overlaid motions. An experimental error analysis is performed and reported.

We present a spatio-temporal analysis of motion at occluding boundaries as an extension of previous results for trans- parent motions. We show how these new results generalize alternative approaches derived in the Fourier domain that are limited by assuming straight occlusion boundaries. Furthermore, we derive a novel hierarchical algorithm that can deal with single, multiple-transparent, and occluded motions.

We describe an algorithm for change detection which is insensitive to both slow and fast temporal variations of scene illumination. Our algorithm is based on statistical decision theory by using a Bayesian approach. The goal is to detect only temporal changes which are induced by true scene changes, like motion, but not changes due to varying illumination or noise. To this end, our algorithm uses a simple illumination model which is invariant to common camera nonlinearities like gamma-nonlinearity. This is combined with a model for the influence of noise as well as an a priori model for the expected properties of the sought change masks. Key ingredients of the resulting algorithm are a suitable test statistic and an adaptive threshold mechanism. As the algorithm can be applied in a noniterative manner, it is also computationally attractive.

Local gray level dependencies of natural images can be modelled by means of co-occurrence matrices containing joint probabilities of gray-level pairs. Texture, however, is a resolution-dependent phenomenon and hence, classification depends on the chosen scale. Since there is no optimal scale for all textures we employ a multiscale approach that acquires textural features at several scales. Thus linear and nonlinear scale-spaces are analyzed by multiscale co-occurrence matrices that describe the statistical behavior of a texture in scale-space. Classification is then performed on the basis of texture features taken from the individual scale with the highest discriminatory power. By considering cross-scale occurrences of gray level pairs, the impact of filters on the feature is described and used for classification of natural textures. This novel method was found to improve classification rates of the common co-occurrence matrix approach on standard textures significantly.

We propose a new real-valued lapped transform for 2D-signal and image processing. Lapped transforms are particularly useful in block-based processing, since their intrinsically overlapping basis functions reduce or prevent block artifacts. Our transform is derived from the modulated lapped transform (MLT), which, as a real-valued and separable transform like the discrete cosine transform, does not allow to unambiguously identify oriented structures from modulus spectra. This is in marked contrast to the (complex-valued) discrete Fourier transform (DFT). The new lapped transform is real-valued, and at the same time allows unambiguous detection of spatial orientation. Furthermore, a fast algorithm for this transform exists. As an application example, we investigate the transform's performance in spectral approaches to image restoration and enhancement in comparison to the DFT.

Restoration of global motion blur requires the blur parameters (direction and extent) to be known. To this end, spectral approaches based on finding zeros in power spectrum or bispectrum estimates are widely reported. In noisy conditions, however, finding spectral zeros is hardly possible as they are covered by the noise power spectrum. In this paper, we separate the determination of blur direction and blur extent. The advantage is that, once the blur direction is known, it suffices to estimate a slice of the power spectrum or bispectrum in this direction. These one-dimensional estimates are considerably less sensitive to noise than higher-dimensional ones. The blur direction is determined by an inertia-like tensor, which assesses spectral anisotropy caused by the motion blur. This approach is applied to noisy low-dose X-ray images, which are particularly hard to process since the blur degrades the signal but not the relatively strong noise.

An important issue in object-based image coding is the efficient description of image segments having an arbitrary shape. This paper outlines a solution to this task using basic principles of transform image coding that are generalized for the case of arbitrarily shaped image segments. The texture inside each region is successively approximated using two-dimensional (2-D) shape-independent basis functions defined on a rectangle circumscribing the given image segment. The resulting texture description exhibits a high energy compactness and is well suited for low bit rate image coding. Unlike other approaches aimed at segment-oriented image coding, the proposed concept does not couple texture description with contour coding. The computational load is kept low, especially for the decoding process.

This contribution describes a new filtering technique for the reduction of noise in medical X-ray images acquired with very low doses, like fluoroscopy images. In such images, only between 10 and 200 X-ray quanta contribute to each pixel. The resulting Poisson statistic causes a strong deterioration of image quality by quantum noise, which, in the observed images, is signal-dependent and exhibits a lowpass-shaped power spectrum. The central part of our approach is a linear/median hybrid filter, which is well-known for its good detail preservation properties. Noise reduction by this filter, however, depends on the presence or absence of underlying signal slopes, and is limited to a small range of spatial frequencies. Also, like median filters this filter tends to generate streaking artifacts. We show how a combination of linear/median hybrid filtering with a multiscale pyramid avoids these shortcomings, while simultaneously improving noise reduction performance substantially.

Describes an algorithm for noise reduction and enhancement of images which is able to take into account anisotropies of signal as well as of noise. Processing is based on subjecting each image to a block DFT, followed by comparing each observed magnitude coefficient to the expected noise standard deviation for it. Depending on this comparison, each coefficient is attenuated the more, the more likely it is that it contains only noise. In addition, the attenuation is made dependent on whether or not the observed coefficient contributes to an oriented prominent structure within the processed image block. Orientation as well as the distinctness with which it occurs are detected in the spectral domain by an inertia-like matrix. Orientation information is additionally exploited to selectively enhance oriented structures, thus only marginally increasing noise as compared to isotropic enhancement.

The focus of this paper is on prediction of uncovered background in moving video coding. A new algorithm for estimating the background texture of uncovered areas is proposed which exploits spatial correlation in addition to temporal correlation and motion information. The surrounding known background area is analysed and spatially extrapolated using a highly efficient spectral domain extrapolation algorithm. The method avoids inherent insufficiencies of conventional static background memories such as strong sensitivity to camera zoom and pan and it can be incorporated into any existing motion compensating coding concept. Several prediction results for typical moving image sequences demonstrate the applicability of the proposed concept.>

Describes a method for displacement estimation in stereoscopic images, which is closely coupled with a segmentation of the pictures into homogeneously displaced regions. The technique is driven by a statistical optimization criterion which assesses the quality of the disparity estimate and of the segmentation, thus improving both of these simultaneously. In addition, the optimization criterion explicitly takes occluded areas into consideration. With the additional help of two constraints, this enables the algorithm to locate regions corresponding to occlusions accurately.>

A technique for displacement estimation in stereo image pairs, which is closely coupled with a segmentation of the images into homogeneously displaced regions, is described. An arbitrary initial displacement field is modified by a displacement vector relaxation, evaluating a statistical optimization criterion. Since the displacements are considered vector by vector during the relaxation, the method provides accurate locations of the boundaries between differently displaced regions. The proposed technique derives its power from the fact that the spatial array of displacements-and hence the resulting segmentation-is modeled by a Gibbs-Markov random field.>

We present a three-stage method for segmenting NMR-datasets of the head into 3D-regions corresponding to different brain matter classes, liquid containing structures, cranium, and background. Our technique works from the beginning with 3D-regions, whose internal ’grey’ values as well as shapes are described by stochastic models. The first phase starts by assuming the entire dataset as consisting of only one region, and then recursively extracts those areas which are not compatible with this hypothesis. During this step, special emphasis is given to the problem of accurately locating the region surfaces. In the second stage, a Bayes classifier groups the regions into different categories, like brain matter, liquid, cranium, and background. Classification errors are corrected largely automatically during the third stage by applying simple knowledge about the topological relationships between the classes.

A novel segmentation technique that starts with the whole image being a single region is presented. First, an object detection scheme, which marks those locations where local statistics deviate significantly from the overall statistics, provides location and approximate shapes of the major objects (regions) in the scene. Exact boundaries are subsequently obtained by a contour relaxation algorithm, which includes a general model for typical region shapes. Object detection and contour relaxation are repeated recursively until a stable segmentation result is achieved. Segmentation results are presented.>