Showing papers in "Graphical Models \/graphical Models and Image Processing \/computer Vision, Graphics, and Image Processing in 1985"

A new method for gray-level picture thresholding using the entropy of the histogram

Abstract: Two methods of entropic thresholding proposed by Pun (Signal Process.,2, 1980, 223–237;Comput. Graphics Image Process.16, 1981, 210–239) have been carefully and critically examined. A new method with a sound theoretical foundation is proposed. Examples are given on a number of real and artifically generated histograms.

Topological Structural Analysis of Digitized Binary Images by Border Following

Abstract: Two border following algorithms are proposed for the topological analysis of digitized binary images. The first one determines the surroundness relations among the borders of a binary image. Since the outer borders and the hole borders have a one-to-one correspondence to the connected components of 1-pixels and to the holes, respectively, the proposed algorithm yields a representation of a binary image, from which one can extract some sort of features without reconstructing the image. The second algorithm, which is a modified version of the first, follows only the outermost borders (i.e., the outer borders which are not surrounded by holes). These algorithms can be effectively used in component counting, shrinking, and topological structural analysis of binary images, when a sequential digital computer is used.

Preattentive processing in vision

Abstract: Visual analysis appears to be functionally divided between an early preattentive level of processing at which simple features are coded spatially in parallel and a later stage at which focused attention is required to conjoin the separate features into coherent objects. Evidence supporting this dichotomy comes from behavioral studies of visual search, from differences in the ease of texture segregation, from reports of illusory conjunctions when attention is overloaded, from subjects' ability to identify simple features correctly even when they mislocate them, and from the substantial benefit of pre-cuing the location of a relevant item when the task requires that features be conjoined but not when simple features are sufficient. Some further studies of search have revealed a striking asymmetry between several pairs of stimuli which differ in the presence or absence of a single part or property. The asymmetry depends solely on which of the pair is allocated the role of target and which is replicated to form the background items. It suggests that search for the presence of a visual primitive is automatic and parallel, whereas search for the absence of the same feature is serial and requires focused attention. The search asymmetry can be used as an additional diagnostic to help define the functional features extracted by the visual system.

Moment-preserving thresolding: A new approach

Abstract: A new approach to automatic threshold selection using the moment-preserving principle is proposed. The threshold values are computed deterministically in such a way that the moments of an input picture is preserved in the output picture. Experimental results show that the approach can be employed to threshold a given picture into meaningful gray-level classes. The approach is described for global thresholding, but it is applicable to local thresholding as well.

Human image understanding : Recent research and a theory

Abstract: The perceptual recognition of objects is conceptualized to be a process in which the image of the input is segmented at regions of deep concavity into simple volumetric components, such as blocks, cylinders, wedges, and cones. The fundamental assumption of the proposed theory, recognition-by-components (RBC), is that a modest set of components [ N probably ≤ 36] can be derived from contrasts of five readily detectable properties of edges in a 2-dimensional image: curvature, collinearity, symmetry, parallelism, and cotermination. The detection of these properties is generally invariant over viewing position and image quality and consequently allows robust object perception when the image is projected from a novel viewpoint or degraded. RBC thus provides a principled account of the heretofore undecided relation between the classic principles of perceptual organization and pattern recognition: The constraints toward regularization (Pragnanz) characterize not the complete object but the object's components. A principle of componential recovery can account for the major phenomena of object recognition: If an arrangement of two or three primitive components can be recovered from the input, objects can be quickly recognized even when they are occluded, rotated in depth, novel, or extensively degraded. The results from experiments on the perception of briefly presented pictures by human observers provide empirical support for the theory.

Texture analysis Anno 1983

Abstract: In this paper the texture analysis methods being used at present are reviewed. Statistical as well as structural approaches are included and their performances are compared. Concerning the former approach, the gray level difference method, filter mask texture measures, Fourier power spectrum analysis, cooccurrence features, gray level run lengths, autocorrelation features, methods derived from texture models, relative extrema measures, and gray level profiles are discussed. Structural methods which describe texture by its primitives and some placement rules are treated as well. Attention has to be paid to some essential preprocessing steps and to the influence of rotation and scale on the texture analysis methods. Finally the problem of texture segmentation is briefly discussed.

Segment-Based Stereo Matching*

Abstract: Images are 2-dimensional projections of 3-dimensional scenes, therefore depth recovery is a crucial problem in image understanding, with applications in passive navigation, cartography, surveillance, and industrial robotics. Stereo analysis provides a more direct quantitative depth evaluation than techniques such as shape from shading, and its being passive makes it more applicable than active range finding imagery by laser or radar. This paper addresses the subproblem of identifying corresponding points in the two images. The primitives we use are groups of collinear connected edge points called segments, and we base the correspondence on the "minimum differential disparity" criterion. The result of this processing is a sparse array disparity map of the analyzed scene. �9 1985 Academic Press, Inc.

Threshold selection based on a simple image statistic

Abstract: The problem of automatic threshold selection is considered. After a brief review of available techniques, a novel method is proposed. It is based on image statistics which can be computed without histogramming the grey level values of the image. A detailed analysis of the properties of the algorithm is then carried out. The effectiveness of the method is shown on a number of practical examples.

Determination of 3D human body postures from a single view

Abstract: In this paper a method is proposed to recover and interpret the 3D body structures of a person from a single view, provided that (1) at least six feature points on the head and a set of body joints are available on the image plane, and (2) the geometry of head and lengths of body segments formed by joints are known. First of all, the feature points on the head in the head-centered coordinate system and their image projections are used to determine a transformation matrix. Then, the camera position and orientations are extracted from the matrix. Finally, the 3D coordinates of the head points expressed in the camera-centered coordinate system are obtained. Starting from the coordinates of the neck, which is a head feature point, the 3D coordinates of other joints one-by-one are determined under the assumption of the fixed lengths of the body segments. A binary interpretation tree is used to represent the 2n − 1 possible body structures, if a human body has n joints. To determine the final feasible body structures, physical and motion constraints are used to prune the interpretation tree. Formulas and rules required for the tree pruning are formulated. Experiments are used to illustrate the pruning powers of these constraints. In the two cases of input data chosen, a unique or nearly unique solution of the body structure is obtained.

Adaptive filters for digital image noise smoothing: An evaluation*

Abstract: Six adaptive noise filtering algorithms were implemented and evaluated. There are (1) median filtering, (2) K-nearest neighbor averaging, (3) gradient inverse weighted smoothing, (4) sigma filtering, (5) Lee additive and multiplicative filtering, and (6) modified Wallis filtering. For the sake of comparison, the mean filter was also included. All algorithms were tested on noise corrupted copies of a composite image consisting of a uniform field, a bar pattern of periods increasing from 2 to 20 pixels, printed text, and a military tank sitting on desert terrain. In one test, uniformly distributed noise between gray levels of −32 and 32 was added to the composite image and filtered. In a second test, multiplicative Gaussian noise with mean 1.0 and standard deviation 0.25 was introduced, then filtered. A 7×7 pixel processing window was used in all six adaptive algorithms and the mean filter for both tests. An adaptive filter was used iteratively with varying window sizes to demonstrate the success of iterative adaptive smoothing. Filtering results were evaluated from statistics, examination of transects plotted from each filtered bar pattern, and from visual ranking by a group of observers.

Boundary conditions for lightness computation in Mondrian World

Abstract: Land's retinex theory of lightness computation explains how for a “Mondrian World” image, consisting of a number of patches each of uniform reflectance, the reflectances can be computed from an image of that object. Horn has shown that the computation can be realised as a parallel process performed by successive layers of cooperating computational cells, arranged on hexagonal grids. However, the layers will, in practice, be arrays of finite extent and it is shown to be critical that cells on array boundaries behave correctly. The computation is first analysed in continuous terms, expressed as the solution of a differential equation with certain boundary conditions, and proved to be optimal in a certain sense. The finite element method is used to derive a discrete algorithm.

Experiments with the intensity ratio depth sensor

Abstract: This paper discusses a modification of the “plane-of-light” approach to range data acquisition which utilizes the ratio of two intensity images. The modification allows depth determination at each image pixel, avoiding the need to scan the plane of light, requiring only the acquisition and processing of two or three intensity images. The depth equation together with three experimental methods for its calculation are presented. Results of the three sensor implementations are given for test scenes.

A distance metric for multidimensional histograms

Abstract: A metric is defined on the space of multidimensional histograms. Such histograms store in thexth location the number of events with feature vectorx; examples are gray level histograms and co-occurrence matrices of digital images. Given two multidimensional histograms, each is “unfolded” and a minimum distance pairing is performed using a distance metric on the feature vectorsx. The sum of the distances in the minimal pairing is used as the “match distance” between the histograms. This distance is shown to be a metric, and in the one-dimensional case is equal to the absolute difference of the two cumulative distribution functions. Among other applications, it facilitates direct computation of the distance between co-occurrence matrices or between point patterns. The problem of finding a translation to minimize the distance between point patterns is also discussed.

Image space shading of 3-dimensional objects

Abstract: Two-dimensional images of 3D objects require realistic shading to create the illusion of depth. Traditional (object space) shading methods require extra data (normal vectors) to be stored with the object description. When object representations are obtained directly from measured data, these normal vectors may be expensive to compute; if the object is modified interactively, they must be recomputed frequently. To avoid these problems a simple shading method is devised which uses only information available in image space, after coordinates have been transformed, hidden surfaces removed, and a complete pre-image of all objects has been assembled. The method uses both the distance from the light source and the surface orientation as the basis for shading. The theory and its implementation are discussed and shaded images of a number of objects are presented.

Delaunay-based representation of surfaces defined over arbitrarily shaped domains

Abstract: A method for constructing surface models from a discrete set of arbitrarily distributed data is described. The surface is represented as a network of planar, triangular faces with vertices at the data points, which is built up on a Delaunay triangulation of the data point projections on the x−y plane. In the paper, the problem of triangulating arbitrarily shaped domains is considered. A new definition of Delaunay triangulation of a set of points constrained by a polygonal boundary is introduced, and some of its basic properties are briefly discussed. A new incremental algorithm for constructing a Delaunay triangulation of an arbitrarily shaped domain is described. This method is also demonstrated to be well suited to generate surface approximations at predefined levels of accuracy, which are based on significant subsets of selected data points.

Codon constraints on closed 2D shapes

Abstract: Codons are simple primitives for describing plane curves. They thus are primarily image-based descriptors. Yet they have the power to capture important information about the 3D world, such as making part boundaries explicit. The codon description is highly redundant (useful for error-correction). This redundancy can be viewed as a constraint on the number of possible codon strings. For smooth closed strings that represent the bounding contour (silhouette) of many smooth 3D objects, the constraints are so strong that sequences containing 6 elements yield only 33 generic shapes as compared with a possible number of 15,625 combinations.

Early orientation selection: Tangent fields and the dimensionality of their support

Abstract: Orientation selection is the inference of orientation information out of images. It is one of the foundations on which other visual structures are built, since it must precede the formation of contours out of pointillist data and surfaces out of surface markings. We take a differential geometric view in defining orientation selection, and develop algorithms for actually doing it. The goal of these algorithms is formulated in mathematical terms as the inference of a vector field of tangents (to the contours), and the algorithms are studied in both abstract and computational forms. They are formulated as matching problems, and algorithms for solving them are reduced to biologically plausible terms. We show that two different matching problems are necessary, the first for 1-dimensional contours (which we refer to as Type I processes) and second for 2-dimensional flows (or Type II processes). We conjecture that this difference is reflected in the response properties of “simple” and “complex” cells, respectively, and predict several other psychophysical phenomena.

Discontinuity detection for visual surface reconstruction

Abstract: A method is described for discontinuity detection in pictorial data It computes at each point a planar approximation of the data and uses the statistics of the differences between the actual values and the approximations for detection of both steps and creases The use of local statistical properties in the residuals provides a detection method that is sensitive to the local context of the data, and avoids the use of arbitrary thresholds The subsequent reconstruction, bounded by the detected discontinuities, avoids Gibbs effects and provides reliable surface measurements

«Form-invariant» topological mapping strategy for 2D shape recognition

Abstract: The basic features of a space-variant sampling in relation to the recognition of 2D shapes are presented The spatial geometry of this sampling structure, common to most biological sensors, is characterized by the presence of a highly sampled region (called fovea) surrounded by an area where the spatial density of the photosensitive elements decreases proportionally to the distance from the central area Two features seem potentially useful for shape recognition: (i) the geometry of the sampling structure can be designed so as to optimally sample objects of different sizes (or equivalently observed from different distances); (ii) the scanning of the sampling points can be performed so as to produce a conformal topological transformation of the input image which is, under achievable conditions, invariant to rotations and scalings of the objects Following on from these observations a software implementation of a visual sensor is presented, allowing—along with a template-matching algorithm—for the recognition of bidimensional shapes independently from their position on the visual field, their spatial orientation, and their distance from the sensing device A comparison with traditional spatially-uniform sampling structures is also presented in order to define the best structure for a given recognition task

Sequential synthesis of natural textures

Abstract: A new method for the generation of natural textures is presented. Using a priori-given second-order statistics (second-order spatial averages or autocorrelation parameters) of a natural texture field as input, a procedure is given to synthesize an artificial texture field in such a way that its second-order statistics are equal to the desired ones. The synthesis is achieved directly without inventing higher order statistics, as was in earlier publications. This method allows us to synthesize gray tone texture fields while “controlling” their second-order statistics in rather large neighborhoods. The synthesized textures are very similar visually to the original natural textures used to compute the second-order statistics, but second-order spatial averages give better results than autocorrelation parameters. This seems to strongly support the conjecture that the visual system is only sensitive to the second-order spatial averages of a given texture field, so that these statistics should be used to model textures.

An image processing architecture for real time generation of scale and rotation invariant patterns

Abstract: An architecture is presented which accomplishes in real time and without computation the complex logarithmic coordinate mapping of an n by n input image Simulation results document the architecture's ability to convert scale changes and rotations of the input image into respectively constant horizontal and cyclic vertical shifts of an output image Because of this property, the output patterns are called shift and rotation invariant Pattern recognition can be performed on the invariant output patterns using translation-invariant methods The invariant output patterns are furthermore shown to possess an intensity preserving property for arbitrary variations in image size The underlying reason for that invariance is due to a new mapping rule which excludes the jacobian factor normally used in a change of coordinates The sensitivity of the architecture to placement of the input image, center mismatch, is discussed in terms of the difference between coordinate transformation and functional transformation methods Simulation results are presented which demonstrate the relative insensitivity of the architecture to center mismatch as opposed to the sensitive nature of cartesian to polar mappings

Efficient diagram understanding with characteristic pattern detection

Abstract: A connection diagram understanding system using a facsimile as its input device has been designed and implemented. The principle of this system is described and examples of its application to some hand-sketched diagrams are shown. In order to reduce processing time, the procedure is started by only accessing the pixels located on the borders of certain meshes to detect characteristic patterns and make a control map. Background judgment and long straight line segment extraction are executed on the control map. Other complicated areas, which are usually a small portion of the whole diagram area, are also indicated by special labels on the map and then processed by a detailed procedure which scans every pixel at these areas. Graph descriptions of diagrams are employed at different steps of the hierarchical understanding. Problems of data compression, diagram retrieval, and diagram editing are discussed.

A theory of binary digital pictures

Abstract: We study 2- and 3-dimensional digital geometry in the context of almost arbitrary adjacency relations. (Previous authors have based their work on particular adjacency relations.) We define a binary digital picture to be a pair whose components are a set of lattice-points and an adjacency relation on the whole lattice. We show how a wide class of digital pictures have natural "continuous analogs." This enables us to use methods of continuous topology in studying digital pictures. We are able to prove general results on the connectivity of digital borders, which generalize results that have appeared in the literature. In the 3-dimensional case we consider the possibility of using a uniform relation on the whole lattice. (In the past authors have used different types of adjacency for "object" and "background.") �9 1985 Academic Press, Inc. PREREQUISITES

Early vision: from computational structure to algorithms and parallel hardware

Abstract: I review a new theoretical framework that from the computational nature of early vision leads to algorithms for solving them and suggests a specific class of appropriate hardware. The common computational structure of many early vision problems is that they are mathematically ill-posed in the sense of Hadamard. Standard regularization analysis can be used to solve them in terms of variational principles that enforce constraints derived from a physical analysis of the problem, see T. Poggio and V. Torre (Artificial Intelligence Lab. Memo No. 773, MIT, Cambridge, Mass., 1984). Studies of human perception may reveal whether some principles of a similar type are exploited by biological vision. It can also be shown that the corresponding variational principles are implemented in a natural way by analog networks, see T. Poggio and C. Koch (Artificial Intelligence Lab. Memo No. 783, MIT, Cambridge, Mass., 1984). Specific electrical and chemical networks for localizing edges and computing visual motion are derived. These results suggest that local circuits of neurons may exploit this unconventional model of computation.

Intelligible encoding of ASL image sequences at extremely low information rates

Abstract: American Sign Language (ASL) is a gestural language used by the hearing impaired. This paper describes experimental tests with deaf subjects that compared the most effective known methods of creating extremely compressed ASL images. The minimum requirements for intelligibility were determined for three basically different kinds of transformations: (1) gray-scale transformations that subsample the images in space and time; (2) two-level intensity quantization that converts the gray scale image into a black-and-white approximation; (3) transformations that convert the images into black and white outline drawings (cartoons). In Experiment 1, five subjects made quality ratings of 81 kinds of images that varied in spatial resolution, frame rate, and type of transformation. The most promising image size was 96 × 64 pixels (height × width). The 17 most promising image transformations were selected for formal intelligibility testing: 38 deaf subjects viewed 87 ASL sequences 1–2 s long of each transformation. The most effective code for gray-scale images is an analog raster code, which can produce images with 0.86 normalized intelligibility (I) at a bandwidth of 2,880 Hz and therefore is transmittable on ordinary 3 KHz telephone circuits. For the binary images, a number of coding schemes are described and compared, the most efficient being an extension of the quadtree method, here termed binquad coding which yielded I = 0.68 at 7,500 bits per second (bps). For cartoons, an even more efficient polygonal transformation with a victorgraph code yielding, for connected straight line segments, is proposed, together with a vectorgraph code yielding, for example, I = 0.56 at 3,900 bps and I = 0.70 at 6,000 bps. Polygonally transformed cartoons offer the possibility of telephonic ASL communication at 4,800 bps. Several combinations of binary image transformations and encoding schemes offer I > 80% at 9,600 bps.

Tracing planar surface motion from a projection without knowing the correspondence

Abstract: The 3-dimensional motion of a planar surface is detected only from the motion of its projected 2-dimensional contour image on the plane of vision. There is no need to know the correspondence of points. The motion is explicitly given by measuring “diameters” of the image contour on the plane of vision. No iterative or matching processes are involved. Numerical examples are also given.

Two-dimensional run-encoding for quadtree representation

Abstract: Two-dimensional run-encoding (2DRE) is a new file structuring method for quadtree representation which does not use explicit pointers. The structure is based on the assignment of linear keys to each cell in an image, and uses these keys to encode maximal leaves. It is shown that storage requirements for such a structure can be reduced by run-encoding the leaves on the value information. Values are extended from the binary encoding of most previous quadtree research, to multivalue (multicolored) encoding, offering greater flexibility, particularly in the field of geographic information processing. Methods for encoding, decoding, search, and overlay (union, intersection) are presented, and shown to be similar to those acting on other quadtree structures in terms of time complexity. The space-efficiency of 2DRE files is similar or superior to other forms of quadtrees.

Image representation using Voronoi tessellation

Abstract: The cells of the Voronoi tessellation are used as primitives to represent image regions. The tessellation is derived from a Poisson point process. The random shapes of cells make the representation attractive for secure transmission.

Aspects of perception and computation in color vision

Abstract: One of the areas which has received little attention in computer vision research is color vision. The main reason for this was technological: color digitizers and monitors were not very common, and the amount of computation and memory needed for chromatic image processing were large and expensive. This is not longer a major factor. In addition, recent studies suggest that color contributes to other visual processes, and therefore cannot be considered as a mechanism which merely adds beauty to scenes. It should be an integral part of any computer vision system, and justifies a more complex treatment. This paper reviews the main aspects of color vision, which originate from different disciplines of science. Physical concepts regarding light and reflection govern the appearance of objects. The structure of the visual system (mainly in primates) may help the construction of computational models of color vision. Psychophysical experiments are a resource of theories which can be implemented and tested as part of a vision system. The discussion in this paper includes the different facts and problems they pose, some of which are currently being explored.

Rule-based image segmentation: A dynamic control strategy approach

Abstract: The structure and functional aspects of a rule-based image segmentation system are briefly described This paper focuses mainly on the control aspects of the system A set of measurements is defined that evaluates the quality of an image segmentation and thereby determines the control To accomplish this, focus of attention areas are created within the image These are processed in an order that depends on their properties, as reflected by the set of performance measures, computed for each area A dynamic control strategy within each area is also determined based on the individual characteristics of that area One aspect involves determining the spatial order in which the system will process the data entries inside the area In addition, an ordering must be established among all the rules included in the model Dynamic strategy setting is formulated as a fuzzy decision-making problem, whose solution depends on the performance parameters Because the individual characteristics of areas are reflected in different performance measurements, the resulting control strategies will vary from one area to the next In addition, the strategy within each area will vary with time to reflect the changing properties This spatial and temporal updating process is designed to ensure both efficiency and improved output