. Existing methods for grouping edges on the basis of local smoothness measures fail to compute complete contours in natural images: it appears that a stronger global constraint is required. Motivated by growing evidence that the human visual system exploits contour closure for the purposes of perceptual grouping [6, 7, 14, 15, 25], we present an algorithm for computing highly closed bounding contours from images. Unlike previous algorithms [11, 18, 26], no restrictions are placed on the type of structure bounded or its shape. Contours are represented locally by tangent vectors, augmented by image intensity estimates. A Bayesian model is developed for the likelihood that two tangent vectors form contiguous components of the same contour. Based on this model, a sparsely-connected graph is constructed, and the problem of computing closed contours is posed as the computation of shortest-path cycles in this graph. We show that simple tangent cycles can be efficiently computed ...

We investigate a measure of "dominant perceived orientation " that has recently been developed to match the output of a human study involving 40 subjects. The results of this measure are compared with humans analyzing seven "teaser" images to test its effectiveness for finding perceptually dominant orientations. The use of low-level orientation is then applied to a "quick search" problem important in image database applications. Since both pigeons and humans are able to perform coarse classification of certain kinds of scenes, e.g., city from country, without taking time or brainpower to solve the image understanding problem, we conjecture that the collective behavior of low-level textural features such as orientation may be doing most of the work. We demonstrate a simple test of global multiscale orientation for quickly searching a database of vacation photos for likely "city/suburb" shots. The orientation features achieve agreement with human classification in 91 out of 98 of the sce...

This thesis presents a perceptual system for a humanoid robot that integrates abilities such as object localization and recognition with the deeper developmental machinery required to forge those competences out of raw physical experiences. It shows that a robotic platform can build up and maintain a system for object localization, segmentation, and recognition, starting from very little. What the robot starts with is a direct solution to achieving figure/ground separation: it simply `pokes around' in a region of visual ambiguity and watches what happens. If the arm passes through an area, that area is recognized as free space. If the arm collides with an object, causing it to move, the robot can use that motion to segment the object from the background. Once the robot can acquire reliable segmented views of objects, it learns from them, and from then on recognizes and segments those objects without further contact. Both low-level and high-level visual features can also be learned in this way, and examples are presented for both: orientation detection and affordance recognition, respectively.

This paper presents a new measure of phase congruency and shows how it can be calculated through the use of wavelets. The existing theory that has been developed for 1-D signals is extended to allow the calculation of phase congruency in 2-D images. It is shown that, for good localization, it is important to consider the spread of frequencies present at a point of phase congruency. An effective method for identifying and compensating for the level of noise in an image is presented

The Saliency Network proposed by Shashua and Ullman (1988) is a well-known approach to the problem of extracting salient curves from images while performing gap completion. This paper analyzes the Saliency Network. The Saliency Network is attractive for several reasons. First, the network generally prefers long and smooth curves over short or wiggly ones. While computing saliencies, the network also fills in gaps with smooth completions and tolerates noise. Finally, the network is locally connected, and its size is proportional to the size of the image. Nevertheless, our analysis reveals certain weaknesses with the method. In particular, we show cases in which the most salient element does not lie on the perceptually most salient curve. Furthermore, in some cases the saliency measure changes its preferences when curves are scaled uniformly. Also, we show that for certain fragmented curves the measure prefers large gaps over a few small gaps of the same total size. In addition, we analyze the time complexity required by the method. We show that the number of steps required for convergence in serial implementations is quadratic in the size of the network, and in parallel implementations is linear in the size of the network. We discuss problems due to coarse sampling of the range of possible orientations. Finally, we consider the possibility of using the Saliency Network for grouping. We show that the Saliency Network recovers the most salient curve efficiently, but it has problems with identifying any salient curve other than the most salient one.

This paper describes a method for aggregating local edge evidences into coherent pieces of contour. An independent representation of corner and junction features provides suitable stop-conditions for the aggregation process and allows to divide contours into meaningful substrings, right from the beginning. The active role of corner and junction points makes the contours converge onto them and greatly reduces the problems associated with purely edge-based methods. A second stage is concerned with completing established contours across regions that are less well-defined by contrast. The algorithm suggested uses the attributes of established structures (e.g. direction of termination) as well as local orientation and edge evidences to constrain possible completions in a rigorous way.

Image segmentation is an important part in any computer vision framework. However, the transition from local low-level representations to useful structures and relations in the intermediate levels has turned out to be a truly difficult problem. This paper addresses the difficult transition from low-level into intermediate-level vision, where the latter deals with producing a description of image and scene attributes in which more global relations are made explicit. We propose to combine a rich attributed contour representation with very general geometric contour relations. The implemented geometric relations, which are proximity, curvilinearity, parallelism and corner-like relations, allow to handle general man-made objects whose projected surfaces can be described by combinations of the defined relations. The combination of rich image attributes and geometric relations allows to discriminate between strong and weak contour relations. Strong relations require that not only the geometrical constraints are met but also that the contour attributes (e.g. photometric) are in agreement. We describe the approach and show some preliminary results.

The function of vision is to get correct and useful answers about the state of the world. However, given that the state of the world is not uniquely specified by the visual input, the visual system must make good guesses or inferences. Thus, theories of visual system functions will be theories of inference, and we need a language in which theories of inference can be described. Analogous to calculus having a minimum expressiveness required to formulate theories in physics, we argue that the language of Bayesian inference is fundamental to quantitatively describe how reliable answers about the world can be obtained from image patterns. Bayes provides a minimal formalism that can deal with the sophistication and versatility of perception missing from some other approaches. Key missing components include the ability to model uncertainty, probabilistic modeling of pattern synthesis as a necessary prerequisite to understanding pattern inference, the means to handle the complexity of natural images, and the diversity of visual tasks. Most of the formal elements that we describe are not new and have their roots in signal detection theory and ideal observer analysis. We start from there to review and codify principles drawn from recent applications of Bayesian decision theory, Bayes nets and pattern theory to vision. To emphasize the

The purpose of this work is to investigate data--driven shifts of focus of attention within the particular framework of scale--space cells. The framework is motivated by the complexity issues of visual object recognition and includes scale--dependent input data reduction and self--similarity across the scales. We propose that image structures (objects and their features) should be represented with a minimum amount of information. For this purpose we use the wavelet representation of the scale which is approximately one octave below the scale at which an object becomes smoothed out and thus indistinguishable from other objects of the same size. We devise a scheme for selecting the regions containing the objects at these characteristic scales. The scheme provides a measure of how interesting each region is for exploitation by higher--level visual modules. Most of the features selected by the attention module are robust under different transformations. We also study the limits that arise ...

Accurate detection and localization of two-dimensional (2D) image features (or `keypoints ') is important for vision tasks such as structure from motion, stereo matching, and line labeling. 2D image features are ideal for these vision tasks because 2D image features are high in information and yet they occur sparsely in typical images. Several methods for the detection of 2D image features have already been developed. However, it is difficult to assess the performance of these methods because no one has produced an adequate definition of corners that encompasses all types of 2D luminance variations that make up 2D image features. The fact that there does not exist a consensus on the definition of 2D image features is not surprising given the confusion surrounding the definition of 1D image features. The general perception of 1D image features has been that they correspond to `edges' in an image and so are points where the intensity gradient in some direction is a local maximum. The S...