Showing papers on "Kernel (image processing) published in 2001"

Robust registration of 2D and 3D point sets

Abstract: This paper introduces a new method of registering point sets. The registration error is directly minimized using general-purpose non-linear optimization (the Levenberg–Marquardt algorithm). The surprising conclusion of the paper is that this technique is comparable in speed to the special-purpose Iterated Closest Point algorithm, which is most commonly used for this task. Because the routine directly minimizes an energy function, it is easy to extend it to incorporate robust estimation via a Huber kernel, yielding a basin of convergence that is many times wider than existing techniques. Finally, we introduce a data structure for the minimization based on the chamfer distance transform, which yields an algorithm that is both faster and more robust than previously described methods.

Convolution Kernels for Natural Language

Abstract: We describe the application of kernel methods to Natural Language Processing (NLP) problems. In many NLP tasks the objects being modeled are strings, trees, graphs or other discrete structures which require some mechanism to convert them into feature vectors. We describe kernels for various natural language structures, allowing rich, high dimensional representations of these structures. We show how a kernel over trees can be applied to parsing using the voted perceptron algorithm, and we give experimental results on the ATIS corpus of parse trees.

A new benchmark quality solution for the buoyancy-driven cavity by discrete singular convolution

Abstract: This article introduces a high-accuracy discrete singular convolution (DSC) for the numerical simulation of coupled convective heat transfer problems. The problem of a buoyancy-driven cavity is solved by two completely independent numerical procedures. One is a quasi-wavelet-based DSC approach, which uses the regularized Shannon's kernel, while the other is a standard form of the Galerkin finite-element method. The integration of the Navier-Stokes and energy equations is performed by employing velocity correction-based schemes. The entire laminar natural convection range of 10 3 h Ra h 10 8 is numerically simulated by both schemes. The reliability and robustness of the present DSC approach is extensively tested and validated by means of grid sensitivity and convergence studies. As a result, a set of new benchmark quality data is presented. The study emphasizes quantitative, rather than qualitative comparisons.

Kernel machine based learning for multi-view face detection and pose estimation

Abstract: Face images are subject to changes in view and illumination. Such changes cause data distribution to be highly nonlinear and complex in the image space. It is desirable to learn a nonlinear mapping from the image space to a low dimensional space such that the distribution becomes simpler tighter and therefore more predictable for better modeling effaces. In this paper we present a kernel machine based approach for learning such nonlinear mappings. The aim is to provide an effective view-based representation for multi-view face detection and pose estimation. Assuming that the view is partitioned into a number of distinct ranges, one nonlinear view-subspace is learned for each (range of) view from a set of example face images of that view (range), by using kernel principal component analysis (KPCA). Projections of the data onto the view-subspaces are then computed as view-based nonlinear features. Multi-view face detection and pose estimation are performed by classifying a face into one of the facial views or into the nonface class, by using a multi-class kernel support vector classifier (KSVC). Experimental results show that fusion of evidences from multi-views can produce better results than using the result from a single view; and that our approach yields high detection and low false alarm rates in face detection and good accuracy in pose estimation, in comparison with the linear counterpart composed of linear principal component analysis (PCA) feature extraction and Fisher linear discriminant based classification (FLDC).

Fast convolution on the sphere

Abstract: We propose fast, exact and efficient algorithms for the convolution of two arbitrary functions on the sphere which speed up computations by a factor $\mathcal{O}(\sqrt{N})$ compared to present methods where N is the number of pixels. No simplifying assumptions are made other than band limitation. This reduces typical computation times for convolving the full sky with the asymmetric beam pattern of a megapixel cosmic microwave background (CMB) mission from months to minutes. Our methods enable realistic simulation and careful analysis of data from such missions, taking into account the effects of asymmetric ``point spread functions'' and far side lobes of the physical beam. While motivated by CMB studies, our methods are general and hence applicable to the convolution or filtering of any scalar field on the sphere with an arbitrary, asymmetric kernel. We show in an Appendix that the same ideas can be applied to the inverse problems of map-making and beam reconstruction by similarly accelerating the transpose convolution which is needed for the iterative solution of the normal equations.

Diffusions and Confusions in Signal and Image Processing

Abstract: In this paper we link, through simple examples, between three basic approaches for signal and image denoising and segmentation: (1) PDE axiomatics, (2) energy minimization and (3) adaptive filtering. We show the relation between PDE's that are derived from a master energy functional, i.e. the Polyakov harmonic action, and non-linear filters of robust statistics. This relation gives a simple and intuitive way of understanding geometric differential filters like the Beltrami flow. The relation between PDE's and filters is mediated through the short time kernel.

Method and apparatus for fast aerial image simulation

Abstract: The present invention provides a method and apparatus for simulating an aerial image projected from an optical system, wherein the optical system includes a pupil and a mask. In general, the method comprises the steps of obtaining parameters for the optical system, calculating a kernel based on an orthogonal pupil projection of the parameters of the optical system onto a basis set, obtaining parameters of the mask, calculating a vector based on an orthogonal mask projection of the parameters of the mask onto a basis set, calculating a field intensity distribution using the kernel and the vector, and obtaining aerial image data from the field intensity distribution.

Image retrieval with relevance feedback: from heuristic weight adjustment to optimal learning methods

Abstract: Various relevance feedback algorithms have been proposed in recent years in the area of content-based image retrieval. This paper gives a brief review and analysis on existing techniques-from early heuristic-based feature weighting schemes to recently proposed optimal learning algorithms. In addition, the kernel-based biased discriminant analysis (KBDA) is proposed to fit the unique nature of relevance feedback as a biased classification problem. As a novel variant of traditional discriminant analysis, the proposed algorithm provides a trade-off between discriminant transform and regression. The kernel form is derived to deal with non-linearity in an elegant way. Experimental results indicate that significant improvement in retrieval performance is achieved by the new scheme.

Structure and motion from uncalibrated catadioptric views

Abstract: In this paper we present a new algorithm for structure from motion from point correspondences in images taken from uncalibrated catadioptric cameras with parabolic mirrors. We assume that the unknown intrinsic parameters are three: the combined focal length of the mirror and lens and the intersection of the optical axis with the image. We introduce a new representation for images of points and lines in catadioptric images which we call the circle space. This circle space includes imaginary circles, one of which is the image of the absolute conic. We formulate the epipolar constraint in this space and establish a new 4/spl times/4 catadioptric fundamental matrix. We show that the image of the absolute conic belongs to the kernel of this matrix. This enables us to prove that Euclidean reconstruction is feasible from two views with constant parameters and from three views with varying parameters. In both cases, it is one less than the number of views necessary with perspective cameras.

Efficient non-parametric adaptive color modeling using fast Gauss transform

Abstract: Modeling the color distribution of a homogeneous region is used extensively for object tracking and recognition applications. The color distribution of an object represents a feature that is robust to partial occlusion, scaling and object deformation. A variety of parametric and non-parametric statistical techniques have been used to model color distributions. In this paper we present a non-parametric color modeling approach based on kernel density estimation as well as a computational framework for efficient density estimation. Theoretically, our approach is general since kernel density estimators can converge to any density shape with sufficient samples. Therefore, this approach is suitable to model the color distribution of regions with patterns and mixture of colors. Since kernel density estimation techniques are computationally expensive, the paper introduces the use of the fast Gauss transform for efficient computation of the color densities. We show that this approach can be used successfully for color-based segmentation of body parts as well as segmentation of many people under occlusion.

Method for enhancing compressibility and visual quality of scanned document images

Abstract: A system and method of image processing for smoothing, denoising, despeckling and sharpening scanned document images which is performed prior to a compression. The scanned image is selectively smoothed by anisotropic diffusion filtering in a single iteration with a 3x3 kernel, which provides denoising, edge-preserving smoothing. The smoothed image data is then selectively sharpened using variable contrast mapping that provides overshoot-free variable-sharpening and despeckling. Image quality is improved while increasing compressibility of the image.

Direct reconstruction of non-Cartesian k-space data using a nonuniform fast Fourier transform

Abstract: An algorithm of Dutt and Rokhlin (SIAM J Sci Comput 1993;14:1368–1383) for the computation of a fast Fourier transform (FFT) of nonuniformly-spaced data samples has been extended to two dimensions for application to MRI image reconstruction. The 2D nonuniform or generalized FFT (GFFT) was applied to the reconstruction of simulated MRI data collected on radially oriented sinusoidal excursions in k-space (ROSE) and spiral k-space trajectories. The GFFT was compared to conventional Kaiser-Bessel kernel convolution regridding reconstruction in terms of image reconstruction quality and speed of computation. Images reconstructed with the GFFT were similar in quality to the Kaiser-Bessel kernel reconstructions for 2562 pixel image reconstructions, and were more accurate for smaller 642 pixel image reconstructions. Close inspection of the GFFT reveals it to be equivalent to a convolution regridding method with a Gaussian kernel. The Gaussian kernel had been dismissed in earlier literature as nonoptimal compared to the Kaiser-Bessel kernel, but a theorem for the GFFT, bounding the approximation error, and the results of the numerical experiments presented here show that this dismissal was based on a nonoptimal selection of Gaussian function. Magn Reson Med 45:908–915, 2001. © 2001 Wiley-Liss, Inc.

A rapid look-up table method for reconstructing MR images from arbitrary K-space trajectories

Abstract: Look-up tables (LUTs) are a common method for increasing the speed of many algorithms. Their use can be extended to the reconstruction of nonuniformly sampled k-space data using either a discrete Fourier transform (DFT) algorithm or a convolution-based gridding algorithm. A table for the DFT would be precalculated arrays of weights describing how each data point affects all of image space. A table for a convolution-based gridding operation would be a precalculated table of weights describing how each data point affects a small k-space neighborhood. These LUT methods were implemented in C++ on a modest personal computer system; they allowed a radial k-space acquisition sequence, consisting of 180 view's of 256 points each, to be gridded in 36.2 ms, or, in approximately 800 ns/point. By comparison, a similar implementation of the gridding operation, without LUTs, required 45 times longer (1639.2 ms) to grid the same data. This was possible even while using a 4/spl times/4 Kaiser-Bessel convolution kernel, which is larger than typically used. These table-based computations will allow real time reconstruction in the future and can currently be run concurrently with the acquisition allowing for completely real-time gridding.

System and method for preparing software for execution in a dynamically configurable hardware environment

Abstract: A system and method for creating run time executables in a configurable processing element array is disclosed. This system and method includes the step of partitioning a processing element array into a number of defined sets of hardware accelerators, which in one embodiment are processing elements called "bins". The system and method then involves decomposing a program desription in object code form into a plurality of "kernel sections", where the kernel sections are defined as those sections of object code which are candidates for hardware acceleration. Next, mapping the identified kernel sections into a number of hardware dependent designs is performed. Finally, a matrix of the bins and the designs is formed for use by the run time system.

Shape deformation: SVM regression and application to medical image segmentation

Abstract: This paper presents a novel landmark-based shape deformation method. This method effectively solves two problems inherent in landmark-based shape deformation: (a) identification of landmark points from a given input image, and (b) regularized deformation the shape of an an object defined in a template. The second problem is solved using a new constrained support vector machine (SVM) regression technique, in which a thin-plate kernel is utilized to provide non-rigid shape deformations. This method offers several advantages over existing landmark-based methods. First, it has a unique capability to detect and use multiple candidate landmark points in an input image to improve landmark detection. Second, it can handle the case of missing landmarks, which often arises in dealing with occluded images. We have applied the proposed method to extract the scalp contours from brain cryosection images with very encouraging results.

Iteratively building displacement maps for image warping

Abstract: A method and apparatus implement techniques for warping a digital image. The techniques select a plurality of elements from a dense mesh (405), each element corresponding to one or more pixels of the image and representing a vector of displacement values. A kernel is applied iteratively to the mesh (411) in order to update the selected elements with new displacement values until a termination condition is achieved. The kernel is selected so that the iterations converge to a solution of an appropriate differential equation. The resultant mesh is applied to the image, thereby warping the image according to the new displacement values (413). The displacement values of the selected elements and the non-selected elements are used as inputs to the kernel. The new displacement values generated by the kernel are written to the selected elements without updating the displacement values of the non-selected elements (409).

Improving an Algorithm for Approximate Pattern Matching

Abstract: We study a recent algorithm for fast on-line approximate string matching. This is the problem of searching a pattern in a text allowing errors in the pattern or in the text. The algorithm is based on a very fast kernel which is able to search short patterns using a nondeterministic finite automaton, which is simulated using bit-parallelism. A number of techniques to extend this kernel for longer patterns are presented in that work. However, the techniques can be integrated in many ways and the optimal interplay among them is by no means obvious. The solution to this problem starts at a very low level, by obtaining basic probabilistic information about the problem which was not previously known, and ends integrating analytical results with empirical data to obtain the optimal heuristic. The conclusions obtained via analysis are experimentally confirmed. We also improve many of the techniques and obtain a combined heuristic which is faster than the original work. This work shows an excellent example of a complex and theoretical analysis of algorithms used for design and for practical algorithm engineering, instead of the common practice of first designing an algorithm and then analyzing it.

Methods and apparatuses for handling single-user applications in multi-user computing environments

Abstract: Methods and apparatuses are provided for controlling application software while switching between sessions in a multi-session computing environment. An apparatus includes memory coupled to switching logic and application program managing logic. The switching logic is configured to selectively switch console control of a computing device between at least two user kernel sessions that are maintained in the memory. The application program managing logic is configured to selectively control at least one application program that is operatively configured within at least one of the user kernel sessions. For example, the application program managing logic can be configured to stop the operation, re-start certain application programs, notify application programs about switching events, and/or adjust the playback of audio and/or video signals associated certain application programs.

The flash-lag effect as a spatiotemporal correlation structure.

Abstract: The flash-lag effect refers to the phenomenon in which a flash adjacent to a continuously moving object is perceived to lag behind it. Phenomenally, the flash appears to be spatially shifted relative to the moving stimulus, and the amount of lag has often been quantified as the flash's nulling position, which is the physical spatial offset needed to establish perceptual alignment. The present study offers a better way to summarize flash-lag data. Instead of plotting data in terms of space, the psychometric function of the observer's relative-position judgment is drawn on spatiotemporal plot. The psychological process underlying illusory lag is formulated as spatiotemporal bias and uncertainty and their estimate as a spatiotemporal convolution kernel that best explains the spatiotemporal psychometric function. Two empirical procedures of kernel estimation are described. One procedure is to fit the free parameters of the kernel to experimental data for continuous motion trajectory. The second is to give an analytical solution to the kernel using experimental data for random motion trajectory. The two procedures yield similar kernels, with negligible spatial bias and uncertainty and substantial temporal bias and uncertainty. In addition, it is demonstrated that an experimental manipulation of temporal predictability of the flash can change the temporal bias in the estimated kernel. The results of this novel analysis reveal that the flash-lag effect is viewed as a spatiotemporal correlation structure, which is largely characterized by the tendency to compare the position of the flash in the past with the position of the moving item in the present.

Generating resized images using ripple free image filtering

Abstract: An optimal filter kernel, formed by convolving a box filter with a filter of fixed integer width and unity area, is used to perform image resizing and reconstruction. The optimal filter has forced zeros at locations along a frequency scale corresponding to the reciprocal of the spacing of one or more pixels that comprise a source image to be resized. When a rescale value for a source image is selected, the optimal filter kernel is computed, mapped to the source image, and centered upon a location within the source image corresponding to the position of an output pixel to be generated. The number of pixels that lie underneath the optimal filter kernel is established by multiplying the number of pixels that comprise the width of the source image by the selected rescale value. Upon mapping the optimal filter kernel, the output pixel values that comprise the resized image are then evaluated by processing the one or more source image pixels, such as through interpolation. Alternatively, the output pixel values of the resized image are calculated by performing partial integral analysis with respect to a standard filter kernel of fixed width and unity area. The output pixel values are calculated by multiplying the pixel value for each pixel under the kernel by the area of the standard filter kernel surrounding the pixel. The products are then summed to reveal the output pixel value, and placed into the output image buffer. Both of these methods speed up the computation process, while producing a ripple free output image.

Constructing facial identity surfaces in a nonlinear discriminating space

Abstract: Recognising face with large pose variation is more challenging than that in a fixed view, e.g. frontal-view, due to the severe non-linearity caused by rotation in depth, self-shading and self-occlusion. To address this problem, a multi-view dynamic face model is designed to extract the shape-and-pose-free facial texture patterns from multi-view face images. Kernel Discriminant Analysis is developed to extract the significant non-linear discriminating features which maximise the between-class variance and minimise the within-class variance. By using the kernel technique, this process is equivalent to a Linear Discriminant Analysis in a high-dimensional feature space which can be solved conveniently. The identity surfaces are then constructed from these non-linear discriminating features. Face recognition can be performed dynamically from an image sequence by matching an object trajectory and model trajectories on the identity surfaces.

Advanced technique for three-dimensional visualization of compound distributions in a rice kernel.

Abstract: A three-dimensional (3D) visualization technique for the compound distribution in a rice kernel was developed. This technique is a combination of sectioning, staining, and digital image postprocessing. By using a special microtome system with adhesive tapes, a set of sequential sections of a rice kernel, which can be preserved with their own set of relative position data, was obtained. A single set of sequential sections was stained by various chemical techniques for the visualization of protein, starch, or lipid content. Each stained section was digitally captured using a CCD imaging device. As the stained areas represent areas containing dye-target complex, the distribution of each compound in the section was visualized in two dimensions. The digitally captured images of a single set of sequential sections were reconstructed to produce a 3D plotting image. As a result, the distributions of various compounds in a rice kernel could be visualized in a new 3D model.