The human visual system is proficient in perceiving three-dimensional shape from the shading patterns in a two-dimensional image. How it does this is not well understood and continues to be a question of fundamental and practical interest. In this paper we present a new quantitative approach to shape-from-shading that may provide some answers. We suggest that the brain, through evolution or prior experience, has discovered that objects can be classified into lower-dimensional object-classes as to their shape. Extraction of shape from shading is then equivalent to the much simpler problem of parameter estimation in a low dimensional space. We carry out this proposal for an important class of 3D objects; human heads. From an ensemble of several hundred laser-scanned 3D heads, we use principal component analysis to derive a low-dimensional parameterization of head shape space. An algorithm for solving shape-from-shading using this representation is presented. It works well even on real im...

Many researchers interested in connectionist models accept that such models are "neurally inspired " but do not worry too much about whether their models are biologically realistic. While such a position may be perfectly justifiable, the present paper attempts to illustrate how biological information can be used to constrain connectionist models. Two particular areas are discussed. The first section deals with visual information processing in the primate and human visual system. It is argued that speed with which visual information is processed imposes major constraints on the architecture and operation of the visual system. In particular, it seems that a great deal of processing must depend on a single bottum-up pass. The second section deals with biological aspects of learning algorithms. It is argued that although there is good evidence for certain coactivation related synaptic modification schemes, other learning mechanisms, including back-propagation, are not currently supported by experimental data.

This thesis describes a connectionist model which learns to perceive spatial events and relations in simple movies of 2-dimensional objects, so as to name the events and relations as a speaker of a particular natural language would. Thus, the model learns perceptually grounded semantics for natural language spatial terms. Natural languages differ -- sometimes dramatically -- in the ways in which they structure space. The aim here has been to have the model be able to perform this learning task for terms from any natural language, and to have learning take place in the absence of explicit negative evidence, in order to rule out ad hoc solutions and to approximate the conditions under which children learn. The central focus of this thesis is a...

The seemingly effortless ability to perceive meaningful objects in an integrated scene actually depends on complex visual processes. The `binding problem' concerns the way in which we select and integrate the separate features of objects in the correct combinations. Experiments suggest that attention plays a central role in solving this problem. Some neurological patients show a dramatic breakdown in the ability to see several objects; their deficits suggest a role for the parietal cortex inthe binding process. However, indirect measures of priming and interference suggest that more information may be implicitly available than we can consciously access.

Hippocampal participation in classical conditioning is described in terms of a multilayer network that portrays stimulus configuration. The network (a) describes behavior in real time, (b) incorporates a layer of "hidden " units positioned between input and output units, (c) includes inputs that are connected to the output directly as well as indirectly through the hidden-unit layer, and (d) uses a biologically plausible backpropagation procedure to train the hidden-unit layer. Nodes and connections in the neural network are mapped onto regional cerebellar, cortical, and hippocampal circuits, and the effect of lesions of different brain regions is formally studied. Computer simulations of the following classical conditioning paradigms are presented: acquisition of delay and trace conditioning, extinction, acquisition-extinction series of delay conditioning, blocking, overshadowing, discrimination acquisition, discrimination reversal, feature-positive discrimination, conditioned inhibition, negative patterning, positive patterning, and generalization. The model correctly describes the effect of hippocampal and cortical lesions in many of these paradigms, as well as neural activity in hippocampus and medial septum during classical conditioning. Some of these results might be extended to the description of anterograde amnesia in human patients. In spite of the vast amount of behavioral and physiological

The processing of spatial information by the visual system shows a number of similarities to the wavelet transforms that have become popular in applied mathematics. Over the last decade, a range of studies has focused on the question of ‘why ’ the visual system would evolve this strategy of coding spatial information. One such approach has focused on the relationship between the visual code and the statistics of natural scenes under the assumption that the visual system has evolved this strategy as a means of optimizing the representation of its visual environment. This paper reviews some of this literature and looks at some of the statistical properties of natural scenes that allow this code to be efficient. It is argued that such wavelet codes are efficient because they increase the independence of the vectors ’ outputs (i.e. they increase the independence of the responses of the visual neurons) by finding the sparse structure available in the input. Studies with neural networks that attempt to maximize the ‘sparsity ’ of the representation have been shown to produce vectors (neural receptive fields) that have many of the properties of a wavelet representation. It is argued that the visual environment has the appropriate sparse structure to make this sparse output possible. It is argued that these sparse/independent representations make it computationally easier to detect and represent the higher-order structure present in complex environmental data.

The extraction of three-dimensional shape from shading is one of the most perceptually compelling, yet poorly understood, aspects of visual perception. In this paper, we report several new experiments on the manner in which the perception of shape from shading interacts with other visual processes such as perceptual grouping, preattentive search (“pop-out”), and motion perception. Our specific findings are as follows: (1) The extraction of shape from shading information incorporates at least two “assumptions ” or constraints—first,that there is a single light source illuminating the whole scene, and second, that the light is shining from “above ” in relation to retinal coordinates. (2) Tokens definedby shading can serve as a basis for perceptual grouping and segregation. (3) Reaction timefor detecting a single convex shape does not increase withthe number of items in the display. This “pop-out ” effect must be based on shading rather than on differences in luminancepolarity, since neither left-right differences nor step changes in luminance resulted in pop-out. (4) When the subjects were experienced, there were no search asymmetries for convex as opposed to concave tokens, but when the subjects were naive, cavities were much easier to detect than convex shapes. (5) The extraction of shape from shading can also provide an input to motion perception. And finally, (6)the assumption of “overhead illumination ” that

Recentreports ofrapid visual search for some feature conjunctions-suggestedthat preattentive vision might be sensitive to scene-based as well as to image-based features (Enns & Rensink, 1990a, 1990b). This study examined visual search for targets defined by the direction of a luminance gradient, a conjunction of luminance and relative location that often: corresponds to ob-= ject curvature and direction of lighting in naturalistic scenes. Experiment 1 showed that such search is influenced by several factors, including the type of gradient, the shape of the contour enclosing the gradient, and the background luminance. These factors were variedsystematically in Experiment 2 in a three-dimensionality rating task and in a visual-search task. The factors combined interactively in the rating task, supporting the presence of an emergent property of three-dimensionality. In contrast, each factor contributed only additively tothe speed of the visualsearch task. This is inconsistent with the view that search is guided by specialized detectors for surface curvature or direction of lighting. Rather, it is in keeping with the view that search is governed by a number of “quick and dirty ” processes that are implemented rapidly and inparallel across the visual field. Conventional theories of preattentive vision claim that simple features such as size, orientation, luminance, and motion are registered automatically and in parallel, whereas the serial spotlight of attention is required todetect conjunctions of these features (Beck, Prazdny, & Rosenfeld, 1983; Julesz, 1984; Treisman, 1986). This claim is based on data from visual-search and texture-segmentation tasks: preattentive sensitivity is implicated by visual-search rates that are relatively independent of display size (i.e., less than 10 msec per item) and by textureboundaries that are perceived spontaneously (i.e., within 50-100 msec of display onset). Recently, there havebeen several reports of very rapid search andlor texture segmentation based on complex conjunctions of these features. Ramachandran (1988) showed that it is possible to segment a texture on the basis of the direction of the luminance gradients within circular texture

Introduction Ultimately we must understand how humans and animals are able to learn the complicated tasks they do. An important component of that learning process is the learning of how to form useful categories from sensory data. Thus the focus of this chapter is that of learning to classify--- learning to recognize that particular patterns belong to the same class which is different from the set of classes that represent other patterns. That such learning can be difficult is illustrated by a commonly used two-dimensional vowel dataset taken from Peterson and Barney (1952) 1 , shown in Figure 1. The data represent different utterances of the common vowels of english. As you can easily see, the distributions from different classes overlap making error-free classification impossible and simple clustering non-optimal. Learning algorithms for classification have been the subject of study in the field of pattern recognition since the 1950s. Such algorithms attempt to

The present study investigates the influence of attention modulation on neural tuning functions. It has been shown in experiments that attention modulation alters neural tuning curves. Attention has been considered, at least, to serve to resolve limiting capacities and to increase the sensitivity to attended stimulus, while the exact functions of attention are still under debate. Inspired by recent experimental results on attention modulation, the present study investigates the influence of changes in the height and base rate of the tuning curve on the encoding accuracy, using the Fisher information. Under on assumption of stimulus-conditional 1 independence of neural responses, we derive explicit conditions which determine when the height and base rate should be increased or decreased to improve the encoding accuracy. Notably, a decrease in the tuning height and base rate can improve the encoding accuracy in some cases. Our theoretical results can predict the effective size of attention modulation on the neural population with respect to the encoding accuracy. We discuss how our method can be used to quantitatively evaluate different aspects of attention function. 1