Abstract. In this paper we present a novel class-based segmentation method, which is guided by a stored representation of the shape of objects within a general class (such as horse images). The approach is different from bottom-up segmentation methods that primarily use the continuity of grey-level, texture, and bounding contours. We show that the method leads to markedly improved segmentation results and can deal with significant variation in shape and varying backgrounds. We discuss the relative merits of class-specific and general image-based segmentation methods and suggest how they can be usefully combined. Keywords: Grouping and segmentation; Figure-ground; Top-down processing; Object classification

In this work we show how to combine bottom-up and topdown approaches into a single figure-ground segmentation process. This process provides accurate delineation of object boundaries that cannot be achieved by either the topdown or bottom-up approach alone. The top-down approach uses object representation learned from examples to detect an object in a given input image and provide an approximation to its figure-ground segmentation. The bottomup approach uses image-based criteria to define coherent groups of pixels that are likely to belong together to either the figure or the background part. The combination provides a final segmentation that draws on the relative merits of both approaches: The result is as close as possible to the top-down approximation, but is also constrained by the bottom-up process to be consistent with significant image discontinuities. We construct a global cost function that represents these top-down and bottom-up requirements. We then show how the global minimum of this function can be efficiently found by applying the sum-product algorithm. This algorithm also provides a confidence map that can be used to identify image regions where additional top-down or bottom-up information may further improve the segmentation. Our experiments show that the results derived from the algorithm are superior to results given by a pure top-down or pure bottom-up approach. The scheme has broad applicability, enabling the combined use of a range of existing bottom-up and top-down segmentations. 1.

Abstract not found

Abstract. We describe a new approach for learning to perform classbased segmentation using only unsegmented training examples. As in previous methods, we first use training images to extract fragments that contain common object parts. We then show how these parts can be segmented into their figure and ground regions in an automatic learning process. This is in contrast with previous approaches, which required complete manual segmentation of the objects in the training examples. The figure-ground learning combines top-down and bottom-up processes and proceeds in two stages, an initial approximation followed by iterative refinement. The initial approximation produces figure-ground labeling of individual image fragments using the unsegmented training images. It is based on the fact that on average, points inside the object are covered by more fragments than points outside it. The initial labeling is then improved by an iterative refinement process, which converges in up to three steps. At each step, the figure-ground labeling of individual fragments produces a segmentation of complete objects in the training images, which in turn induce a refined figure-ground labeling of the individual fragments. In this manner, we obtain a scheme that starts from unsegmented training images, learns the figure-ground labeling of image fragments, and then uses this labeling to segment novel images. Our experiments demonstrate that the learned segmentation achieves the same level of accuracy as methods using manual segmentation of training images, producing an automatic and robust top-down segmentation. 1

We construct a Bayesian model that integrates topdown with bottom-up criteria, capitalizing on their relative merits to obtain figure-ground segmentation that is shape-specific and texture invariant. A hierarchy of bottom-up segments in multiple scales is used to construct a prior on all possible figure-ground segmentations of the image. This prior is used by our top-down part to query and detect object parts in the image using stored shape templates. The detected parts are integrated to produce a global approximation for the object’s shape, which is then used by an inference algorithm to produce the final segmentation. Experiments with a large sample of horse and runner images demonstrate strong figure-ground segmentation despite high object and background variability. The segmentations are robust to changes in appearance since the matching component depends on shape criteria alone. The model may be useful for additional visual tasks requiring labeling, such as the segmentation of multiple scene objects. 1.

we act on the world, we care which glass is ours, which object we already have retrieved, and whether all the cows that left the barn in the morning have returned. Object individuation consists of determining the numerically distinct (distinct in the sense of distinct one) objects that articulate a given scene. Studies of object individuation in infancy typically concern the simplest individuation problem: es- tablishing whether one single object or two distinct objects are involved in some event. Adults bring a wide variety of information to bear on the task of object indi- viduation, including spatiotemporal information (one object cannot be in two places at the same time), property information (a red plastic entity seen on one oc- casion is unlikely to be the same individual as a yellow cloth entity seen on an- other), and kind information (a dog cannot be the same individual as a table). Under many circumstances, spatiotemporal information is primary; if we see an enti

The present studies investigate infants reliance on object kind information in solving the problem of object individuation. Two experiments explored whether adults, 10- and 12month -old infants could use their knowledge of ducks and cars to individuate an ambiguous array consisting of a toy duck perched on a toy car into two objects. A third experiment investigated whether 10-month-old infants could use their knowledge of cups and shoes to individuate an array consisting of a cup perched on a shoe into two objects. Ten-month-old infants failed to use object kind information alone to resolve the ambiguity with both pairs of objects. In contrast, infants this age succeeded in using spatiotemporal information to segment the array into two objects, i.e. they succeeded if shown that the duck moved independently relative to the car, or the cup relative to the shoe. Twelve-month-old infants, as well as adults, succeeded at object individuation on the basis of object kind information alone.

ts, like adults, draw upon spatiotemporal information---information about the spatial arrangements and motions of visible surfaces---to establish representations of discrete individuals. Two objects separated in space (on the frontal plane or in depth), or moving on spatiotemporally discontinuous trajectories, are resolved into distinct individuals (e.g., Baillargeon, 1991, 1995; Spelke, 1991; Spelke, von Hofsten, & Kestenbaum, 1989; von Hofsten & Spelke, 1985; Xu & Carey, 1996). In her previous work, Needham has shown that by 4.5 months of age, infants also draw upon featural information to resolve ambiguous displays 55 Journal of Experimental Child Psychology 78, 55--60 (2001) doi:10.1006/jecp.2000.2603, available online at http://www.idealibrary.com on 0022-0965/01 $35.00 Copyright 2001 by Academic Press All rights of reproduction in any form reserved. Address correspondence and reprint requests to Susan Carey, Department of Psychology, New York University, 6 Washington Place, 7t

Robots and humans receive partial, fragmentary hints about the world’s state through their respective sensors. In this paper, we focus on some fundamental problems in perception that have attracted the attention of researchers in both robotics and infant development: object segregation, intermodal inte-gration, and the role of embodiment. We concentrate on identifying points of contact between the two fields, and also important questions identified in one field and not yet addressed in the other. For object segregation, both fields have examined the idea of using “key events ” where perception is in some way simplified and the infant or robot acquires knowledge that can be exploited at other times. We examine this parallel research in some detail. We propose that the identification of the key events themselves constitutes a point of contact between the fields. And although the specific algorithms used in robots are not easy to relate to infant development, the overall “algorithmic skeleton ” formed by the set of algorithms needed to identify and exploit key events may in fact form a basis for mutual dialogue.

Human concept learning depends upon perception. Our concept of “car ” is built out of perceptual features such as “engine, ” “tire, ” and “bumper. ” However, recent research indicates that the dependency works both ways. We see bumpers and engines in part because we have acquired “car ” concepts and detected examples of them. Perception both influences and is influenced by the concepts that we learn. We have been exploring the psychological mechanisms by which concepts and perception mutually influence one another, and building computational models to show that the circle of influences is benign rather than vicious. Perceptual Learning Is “Early ” Neurologically, Functionally, and Developmentally An initial suggestion that concept learning influences perception comes from a consideration of the differences between novices and experts. Experts in many domains, including radiologists, wine tasters, and Olympic judges, develop specialized perceptual tools for analyzing the objects in their domains of expertise. Much of training and expertise involves not only developing a database of cases or explicit strategies for dealing with the world but also tailoring perceptual processes to more efficiently represent the world (Gibson 1991). Tuning one’s perceptual representation to the environment is a risky proposition. Once a perceptual representation has been altered, it affects all “downstream ” processes that act as consumers of this altered representation. It makes sense to adapt perceptual systems slowly and conservatively. However, the payoffs for perceptual flexibility are also too enticing to forego. They allow an organism to respond quickly, efficiently, and effectively to stimuli without dedicating on-line attentional resources. Instead of strategically determining how to use an unbiased perceptual representation to fit one’s needs, it is often easier to rig up a perceptual system to give task-relevant representations, and then simply leave this rigging in place without strategic control. Perceptual learning is early in several senses: neurological, functional, and developmental.