Showing papers in "Graphical Models \/graphical Models and Image Processing \/computer Vision, Graphics, and Image Processing in 1989"

Multiresolution elastic matching

Abstract: Matching of locally variant data to an explicit 3-dimensional pictorial model is developed for X-ray computed tomography scans of the human brain, where the model is a voxel representation of an anatomical human brain atlas. The matching process is 3-dimensional without any preference given to the slicing plane. After global alignment the brain atlas is deformed like a piece of rubber, without tearing or folding. Deformation proceeds step-by-step in a coarse-to-fine strategy, increasing the local similarity and global coherence. The assumption underlying this approach is that all normal brains, at least at a certain level of representation, have the same topological structure, but may differ in shape details. Results show that we can account for these differences.

Digital topology: introduction and survey

Abstract: Digital topology deals with the topological properties of digital images: or, more generally, of discrete arrays in two or more dimensions. It provides the theoretical foundations for important image processing operations such as connected component labeling and counting, border following, contour filling, and thinning—and their generalizations to three- (or higher-) dimensional “images.” This paper reviews the fundamental concepts of digital topology and surveys the major theoretical results in the field. A bibliography of almost 140 references is included.

Automatic thresholding of gray-level pictures using two-dimensional entropy

Abstract: Automatic thresholding of the gray-level values of an image is very useful in automated analysis of morphological images, and it represents the first step in many applications in image understanding. Recently it was shown that by choosing the threshold as the value that maximizes the entropy of the 1-dimensional histogram of an image, one might be able to separate, effectively, the desired objects from the background. This approach, however, does not take into consideration the spatial correlation between the pixels in an image. Thus, the performance might degrade rapidly as the spatial interaction between pixels becomes more dominant than the gray-level values. In this case, it becomes difficult to isolate the object from the background and human interference might be required. This was observed during studies that involved images of the stomach. The objective of this report is to extend the entropy-based thresholding algorithm to the 2-dimensional histogram. In this approach, the gray-level value of each pixel as well as the average value of its immediate neighborhood is studied. Thus, the threshold is a vector and has two entries: the gray level of the pixel and the average gray level of its neighborhood. The vector that maximizes the 2-dimensional entropy is used as the 2-dimensional threshold. This method was then compared to the conventional 1-dimensional entropy-based method. Several images were synthesized and others were obtained from the hospital files that represent images of the stomach of patients. It was found that the proposed approach performs better specially when the signal to noise ratio (SNR) is decreased. Both, as expected, yielded good results when the SNR was high (more than 12 dB).

Texture description and segmentation through fractal geometry

Abstract: Fractal geometry is receiving increased attention as a model for natural phenomena In this paper we first present a new method for estimating the fractal dimension from image surfaces and show that it performs better at describing and segmenting generated fractal sets Since the fractal dimension alone is not sufficient to characterize natural textures, we define a new class of texture measures based on the concept of lacunarity and use them, together with the fractal dimension, to describe and segment natural texture images

Finite topology as applied to image analysis

Abstract: The notion of a cellular complex which is well known in the topology is applied to describe the structure of images. It is shown that the topology of cellular complexes is the only possible topology of finite sets. Under this topology no contradictions or paradoxes arise when defining connected subsets and their boundaries. Ways of encoding images as cellular complexes are discussed. The process of image segmentation is considered as splitting (in the topological sense) a cellular complex into blocks of cells. The notion of a cell list is introduced as a precise and compact data structure for encoding segmented images. Some applications of this data structure to the image analysis are demonstrated.

An analytic solution for the perspective 4-point problem

Abstract: The perspective n-point (PnP) problem is the problem of finding the position and orientation of a camera with respect to a scene object from n correspondence points. The authors propose an analytic solution for the perspective 4-point problem. The solution is found by replacing the four points with a pencil of three lines and by exploring the geometric constraints available with the perspective camera model. The P4P problem is cast into the problem of solving a biquadratic polynomial equation in one unknown.

Contrast enhancement technique based on local detection of edges

Abstract: A digital processing technique is proposed in order to enhance image contrast without significant noise enhancement. The technique, derived from Gordon's algorithm, accounts for visual perception criteria, namely for contour detection. The efficiency of our algorithm is compared to Gordon's and to the classical ones.

A new method for image segmentation

Abstract: In applications involving visual inspection, it is often required to separate objects from background, in conditions of poor and nonuniform illumination In such cases one has to rely on adaptive methods that learn the illumination from the given images and base the object/background decision on this information We here present a new method for image segmentation via adaptive thresholding The threshold surface is determined by interpolating the image gray levels at points where the gradient is high, indicating probable object edges Several methods of data interpolation to levels given at scattered points in the image plane are discussed One method is tested on several examples and the segmentation results are compared to previously proposed adaptive thresholding algorithms

Classification of newspaper image blocks using texture analysis

Abstract: An important step in the analysis of images of printed documents is the classification of segmented blocks into categories such as half-tone photographs, text with large letters, text with small letters, line drawings, etc. In this paper, a method to classify blocks segmented from newspaper images is described. It is assumed that homogeneous rectangular blocks are first segmented out of the image using methods such as run-length smoothing and recursive horizontal/vertical cuts. The classification approach is based on statistical textural features and feature space decision techniques. Two matrices, whose elements are frequency counts of black-white pair run lengths and black-white-black combination run lengths, are used to derive texture information. Three features are extracted from the matrices to determine a feature space in which block classification is accomplished using linear discriminant functions. Experimental results using different block segmentation results, different newspapers, and different image resolutions are given. Performance and speed with different image resolutions are indicated.

Illumination independent change detection for real world image sequences

Abstract: Change detection plays a very important role in many vision applications Most change detection algorithms assume that the illumination on a scene will remain constant Unfortunately, this assumption is not necessarily valid outside a well-controlled laboratory setting The accuracy of existing algorithms diminishes significantly when confronted with image sequences in which the illumination is allowed to vary In this note, we present two techniques for change detection that have been developed to deal with the more general scenario where illuination is not assumed to be constant A detailed description of both new methods, the derivative model method and the shading model method, is provided Results are presented for applying each of the techniques discussed to various image pairs

A simple approach for the estimation of circular arc center and its radius

Abstract: A simple exact solution for estimating the location of the center of a circular arc and its radius is suggested. Its features are demonstrated with the help of a computer program.

A nonlinear laplace operator as edge detector in noisy images

Abstract: An edge detection scheme is developed robust enough to perform well over a wide range of signal-to-noise ratios. It is based upon the detection of zero crossings in the output image of a nonlinear Laplace filter. Specific characterizations of the nonlinear Laplacian are its adaptive orientation to the direction of the gradient and its inherent masks which permit the development of approximately circular (isotropic) filters. We have investigated the relation between the locally optimal filter parameters, smoothing size, and filter size, and the SNR of the image to be processed. A quantitative evaluation shows that our edge detector performs at least as well—and in most cases much better—than edge detectors. At very low signal-to-noise ratios, our edge detector is superior to all others tested.

Expert systems for image processing: knowledge-based composition of image analysis processes

Abstract: Recently several expert systems for image processing were proposed to facilitate the development of image analysis processes. They use the knowledge about image processing techniques to compose complex image analysis processes from primitive image processing operators. In this paper, we classify them into the following four categories and discuss their objectives, knowledge representation, reasoning methods, and future problems: (1) consultation system for image processing, (2) knowledge-based program composition system, (3) rule-based design system for image segmentation algorithms, and (4) goal-directed image segmentation system. In the latter half of the paper, we emphasize the importance of image analysis strategies in realizing effective image analysis: analysis using the pyramid (multi-resolution) data structure, combination of edge-based and region-based analyses, and so on. We propose two methods of representing image analysis strategies: one from a software engineering viewpoint and the other from a knowledge representation viewpoint. Several examples are given to demonstrate the effectiveness of these methods.

Stochastic image pyramids

Abstract: A new class of image pyramids is introduced in which a global sampling structure close to that of the twofold reduced resolution next level is generated exclusively by local processes. The probabilistic algorithm exploits local ordering relations among independent identically distributed random variables. The algorithm is superior to any coin tossing based procudure and converges to an optimal sampling structure in only three steps. It can be applied to either 1- or 2-dimensional lattices. Generation of stochastic pyramids has broad applicability. We discuss in detail curve processing in 2-dimensional image pyramids and labeling the mesh in massively parallel computers. We also mention investigation of the robustness of multiresolution algorithms and a fast parallel synthesis method for nonhomogeneous anisotropic random patterns.

Fast surface tracking in three-dimensional binary images

Abstract: In medical 3-dimensional display, an input scene is represented by an array of volume elements (abbreviated as “voxels”), and an object in the scene is specified as a “connected” set of voxels. In such applications, surface tracking is an important, and often time-consuming, precursory step. One of the most efficient surface detection algorithms reported in the literature tracks a specified surface of a 3-dimensional object by visiting each boundary face in the surface twice. In this paper, we present a new definition of discrete objects and boundaries that leads to a modified algorithm, which, on the average, visits only one-third of the boundary faces twice (and the rest once). The algorithm has been implemented in our display software package and is found to achieve a run-time reduction of approximately 35%. This timing includes the computation of surface-normal information which is needed for the realistic rendering of surfaces. Without this computation, the saving would be about 55%.

Extraction of early perceptual structure in dot patterns: integrating region, boundary, and component gestalt

Abstract: This paper presents a computational approach to extracting basic perceptual structure, or the lowest level grouping in dot patterns. The goal is to extract the perceptual segments of dots that group together because of their relative locations. The dots are interpreted as belonging to the interior or the border of a perceptual segment, or being along a perceived curve, or being isolated. To perform the lowest level grouping, first the geometric structure of the dot pattern is represented in terms of certain geometric properties of the Voronoi neighborhoods of the dots. The grouping is accomplished through independent modules that posses narrow expertise for recognition of typical interior dots, border dots, curve dots, and isolated dots, from the properties of the Voronoi neighborhoods. The results of the modules are allowed to influence and change each other so as to result in perceptual components that satisfy global, Gestalt criteria such as border and curve smoothness and component compactness. Such latera; communication among the modules makes feasible a perceptual interpretation of the local structure in a manner that best meets the global expectations. Thus, an integration is performed of multiple constraints, active at different perceptual levels and having different scopes in the dot pattern, to infer the lowest level perceptual structure. The local interpretations as well as lateral corrections are performed through constraint propagation using a probabilistic relaxation process. The result is a partitioning of the dot pattern into different perceptual segments or tokens. Unlike dots, these segments posses size and shape properties in addition to locations.

The Laplacian-of-Gaussian kernel: a formal analysis and design procedure for fast, accurate convolution and full-frame output

Abstract: An LoG of space constant σ can be decomposed into the product of a Gaussian and an LoG mask (Chen, Huertas, and Medioni, IEEE Trans. Pattern Anal. Mach. Intell. PAMI-9, 1987, 584–590). The resulting LoG has space constant σ1

Fractal-based analysis and interpolation of 3D natural surface shapes and their application to Terrain modeling

Abstract: This paper describes the fractal-based analysis and interpolation of 3D natural surface shapes and their application to terrain modeling in a sensor simulation for an earth resource satellite. After introducing the fractional Brownian function as a mathematical model for describing natural fractals and discussing its properties, we present fractal-based techniques for analyzing and interpolating 3D surface shapes. These techniques are essentially based on the approximation of the surface by the fractional Brownian function. Finally, the application of these techniques to terrain modeling is described. The analysis-followed-by-interpolation method is successfully applied for generating the digital terrain map of high spatial resolution from the existing coarse terrain data.

Hough transform for line recognition complexity of evidence accumulation and cluster detection

Abstract: Memory and time requirements of the Hough transform for the detection of lines in binary images are compared for different parametrizations of lines. This is done in terms of accumulator usage which can be treated independently of the parameter quantization and of accumulator size which is compared for the hifi-quantization, i.e., the coarsest quantization so that no two realizable lines in the image correspond to the same accumulator cell. Cluster detection in Hough space is discussed and a new adaptive algorithm is proposed to cure some deficiencies of existing algorithms. Now, the hifi-quantization allows exact determination and comparison of the complexity of both filling the accumulator and detecting clusters.

Automatic recognition and representation of shape-based features in a geometric modeling system

Abstract: A method is presented for the automatic identification and extraction of feature information from the solid model of an object. The procedure consists in recognizing shape features, extracting these features as solid volumes, and arranging them in a hierarchical structure. This hierarchical model, described in this article, represents the main shape of the object at the highest levels of abstraction and its form features at lower levels of specification. The system is divided into three modules: feature recognition, feature extraction and feature organization. The recognitition step works on a face-based representation of solid objects, called a face adjacency hypergraph and it takes advantage of Kyprianou's method ( Shape Classification in Computer-Aided-Design , Ph.D. thesis, Computer Laboratory, University of Cambridge, England, July 1980). In the extraction phase each recognized form feature is completed by dummy entities in order to form a feasible object and in the organization step the completed features are arranged into a hierarchical graph, called Structured Face Adjacency Hypergraph.

Optimal corner detector

Abstract: A corner is defined as the junction point of two or more stright line edges. Corners are special features in a image. They are of great use in computing the optical flow and structure from motion. In this paper, we report an optimal corner detector which uses a mathematical model for a corner. An optimal gray tone corner detector is derived for a restricted case of corners, i.e., corners made by lines which are symmetric about a horizontal axis. The resultant corner detector is described by the product of the sine in x and an exponential in the y direction in a portion of the mask and by the product of two sines in x and y directions in the remaining portion. It is then generalized to include any corner of an arbitrary angle and orientation. This results in an approximation of all corners by a total of twelve major types. It is observed that all the twelve masks can actually be configured with four smaller sub-masks, and this results in a significant reduction in the computations. The computations are further reduced by using the separability of masks. Results for synthetic and real scenes are reported.

A Direct Method for Locating the Focus of Expansion

Abstract: We address the problem of recovering the motion of a monocular observer relative to a rigid scene. We do not make any assumptions about the shapes of the surfaces in the scene, nor do we use estimates of the optical flow or point correspondences. Instead, we exploit the spatial gradient and the time rate of change of brightness over the whole image and explicitly impose the constraint that the surface of an object in the scene must be in front of the camera for it to be imaged.

A new surface interpolation technique for reconstructing 3D objects from serial corss-sections

Abstract: A new method for automatically reconstructing a 3-dimensional object from serial cross sections is presented in this paper. The method combines the techniques of dynamic elastic contour interpolation, spline theory, and quadratic variation-based surface interpolation. In the proposed method, the initial description of the object is formed by applying the elastic interpolation algorithm to generate a series of intermediate contours between each pair of consecutive cross sections. After this, a preliminary processing for surface computation is carried out by mapping the contours into the domain of surface function and then using spline functions to calculate the initial surface values. Based on the output from preliminary processing, we apply the quadratic variation-based surface interpolation algorithm to calculate the final surface representation. Since our method takes the continuity of high order derivatives into consideration, smooth and complete surfaces of a 3D object can thus be reconstructed.

Neighbor finding in images represented by octrees

Abstract: Algorithms are presented for moving between adjacent blocks in an octree representation of an image. Motion is possible in the direction of a face, edge, and a vertex, and between blocks of arbitrary size. The algorithms are based on a generalization and simplification of techniques developed earlier for two dimensions (i.e., in quadtrees). They are also applicable to quadtrees. The difference lies in the graph-theoretical classification of adjacencies—i.e., in terms of vertices, edges, and faces. Algorithms are given for octrees that are implemented with pointers and with pointerless representations such as the linear octree.

Automating knowledge acquisition for aerial image interpretation

Abstract: The interpretation of aerial imagery requires substantial knowledge about the scene under consideration. Knowledge about the type of scene-airport, suburban housing development, urban city-aids in low-level and intermediate level image analysis and will drive high-level interpretation by constraining search for plausible consistent scene models. Collecting and representing large knowledge bases requires specialized tools. In this paper we describe the organization of a set of tools for interactive knowledge acquisition of scene primitives and spatial constraints for interpretation of aerial imagery. These tools include a user interface for interactive knowledge acquisition, an automated knowledge compiler that transforms schemabased assertions into productions that are directly executable by our interpretation system, and a performance analysis tool that generates a critique of the final interpretation. The generality of these tools is demonstrated by the generation of rules for a new task, suburban house scenes, and the analysis of a set of imagery by our interpretation system.

A syntactic geometric approach to recognition of dimensions in engineering machine drawings

Abstract: An algorithm for recognizing dimensions in engineering machine drawings that employs a syntactic/geometric approach along with a specific deterministic finite automation (DFA) is presented and demonstrated. First, the problem of distinguishing object from interpretation lines is addressed. Then, dimension-sets, their components, kinds, and types are defined and illustrated. A DFA called the dimension-set profiler is used to determine the profile of a given dimension-set. The resulting profile is used to obtain a sketch which is parsed by the dimensioning grammar to yield a conceptual web. This web, in turn, is converted into a geometric web by substituting components labeling nodes by their line description. These line descriptions are compared to the lines in the actual dimension-set. A certain degree of redundancy is introduced to ascertain valid recognition.

Binary image algebra and optical cellular logic processor design

Abstract: Techniques for digital optical cellular image processing are presented. A binary image algebra (BIA), built from five elementary images and three fundamental operations, serves as its software and leads to a formal parallel language approach to the design of parallel binary image processing algorithms. Its applications and relationships with other computing theories demonstrate that BIA is a powerful systematic tool for formalizing and analyzing parallel algorithms. Digital optical cellular image processors (DOCIPs), based on cellular automata and cellular logic architectures, serve as its hardware and implement parallel binary image processing tasks efficiently. An algebraic structure provides a link between the algorithms of BIA and architectures of DOCIP. Optical computing suggests an efficient and high-speed implementation of the DOCIP architectures because of its inherent parallelism and 3D global free interconnection capabilities. Finally, the instruction set and the programming of the DOCIPs are illustrated.

Connected component labeling of binary images on a mesh connected massively parallel processor

Abstract: An algorithm for connected component labeling of binary patterns using SIMD mesh connected computers is presented. The algorithm consists of three major steps: identifying exactly one point (seed point) within each connected component (region), assigning a unique label to each seed point, and expanding the labels to fill all pixels in the respective regions. Two approaches are given for identifying seed points. The first approach is based on shrinking and the second on the iterative replacement of equivalent labels with local minima or maxima. The shrinking algorithm reduces simply connected regions into single pixels, but multiply connected regions form rings around the holes contained in the regions. A parallel algorithm is developed to break each such ring at a single point. The broken rings are then reduced to single pixels by reshrinking. With iterations consisting of shrinking, breaking rings, if any, and reshrinking, each pattern (of any complexity) is reduced to isolated points within itself. In the second approach every region pixel in the image is initially given a unique label equal to its address in the image. Every 3 × 3 neighborhood in the image is then examined in parallel to replace the central label with the maximum (or minimum) of the labels assigned to the set of region pixels in the neighborhood. This is done iteratively until there is no further change. The seed points are then the locations where the pixel addresses match their converged labels. A parallel sorting method is used for assigning a consecutive set of numbers as labels to the seed points. Parallel expansion up to the boundaries of the original patterns then completes the connected component labeling. The computational complexities of the algorithm are discussed.

Unsupervised textured image segmentation using feature smoothing probabilistic relaxation techniques

Abstract: This paper describes an unsupervised textured image segmentation algorithm. The goals are to improve textured image segmentation results, especially along the borders of regions; and to take into account the spatial relationship among pixels to improve the segmentation of region interiors. An improved method to extract textured energy features in the feature extraction stage is described. The method is based upon an adaptive noise smoothing concept which takes the nonstationary nature of the problem into account. Texture energy features are first estimated using a window of small size to reduce the possibility of mixing statistics along region borders. The estimated texture energy feature values are then smoothed by a quadrant filtering method to reduce the variability of the estimates while retaining the region border accuracy. In this unsupervised segmentation system, the estimated feature values are used in a K-means clustering algorithm to estimate the class statistics. The estimated class statistics are then used by the Bayes classifier to make an initial probabilistic labeling. The spatial constraints are then enforced through the use of a probabilistic relaxation algorithm. Limiting the probability labels by probability thresholding is used to make the relaxation iteration more efficient. The trade-off between efficiency and degradation of performance is studied. Finally, an overview of the proposed textured image segmentation system is provided and comparisons of overall performance are made.

Fast segmentation of range imagery into planar regions

Abstract: A technique is presented for rapidly dividing surfaces in range imagery into regions satisfying a common homogeneity criterion. The result is a segmentation of the range information into approximately planar surface regions. Key features that enhance that algorithm's speed include the development of appropriate region descriptors and the use of fast region comparison techniques for segmentation decisions. The algorithm is a split-and-merge segmentation approach, where the homogeneity criteria is based on a 3-parameter planar surface description technique. The three parameters are two angles describing the orientation of the normal to the local best fit plane and the original range value. Speed is achieved because both the region splitting and the rejection of merge possibilities can often be based on simple comparisons of only the two orientation parameters. A fast, but more complex region-to-region range continuity test is also developed, for use when the orientation homogeneity tests are inconclusive. The importance of merge ordering is considered, and in particular, an effective ordering technique based on dynamic criteria relaxation is demonstrated. Example segmentations of simple and complex range data images are shown, and the effects of noise and preprocessing are examined.