Visual Form Detection in Three-dimensional Space

W. R. Uttal

• Publisher: Psychology Press
• Year: 2013
• ISBN:
  9781134922536
• Edition: 1
• Topic: Psychology
• Subtopic: History & Theory in Psychology
• Index: Psychology

Published in the year 1982, Visual Form Detection in Three-dimensional Space is a valuable contribution to the field of Cognitive Psychology.

1 Form and Process

INTRODUCTION

How do we gain knowledge about the external world? This is the foundation question of epistemology. How do people visually perceive forms? This is the fundamental question that has guided one branch of psychological thought for centuries (and, incidentally, has directed most of my laboratory research for the last decade). The conceptual similarity of these two questions makes the visual laboratory nothing less than the empirical arm of technical epistemology. One, therefore, has to be awed by the audaciousness, if not the pretentiousness, of what we perceptual psychologists are attempting to do in our laboratories. This is so in spite of the fact that the seemingly silly little experiments we perform sometimes obscure the grandeur of the underlying issues under attack. Make no mistake; what we are attempting is a formidable task even in the light of the powerful new tools and methods of which earlier experimental epistemologists could not have dreamed.

It takes no great historical insight to appreciate that the questions currently asked have been asked by others for millenia. The history of the problem of form perception contains such illustrious (and sometimes unexpected) names as Plato, Aristotle, Democritus, Euclid, Alhazen, Seneca, Galen, Avicenna, Grossteste, Descartes, Da Vinci, Vesalius, Kepler, Locke, Berkeley, Hobbes, Goethe, Müller, Helmholtz, Mach, Wundt, James, Koffka, Wertheimer, and Gibson. A large company of other historical figures, as well as a growing army of our contemporaries have also been concerned with various aspects of the form perception problem. What good company we are in when involved in such a quest! And, how enticing the magnitude of the task makes that quest! Not only is the issue dignified by its antiquity, but it is also broad enough to allow one to pursue almost any kind of scientific activity while remaining within the fold of perceptual psychology. It is possible to fiddle with exotic computers and displays; it is possible to carry out psychophysical experiments; it is possible to manipulate mathematical concepts and logical simulations; it is possible to record electrophysiological correlates; it is even possible to become what Jerry Fodor has called a speculative psychologist and manipulate nothing other than a pen or a typewriter, the enormous published data base, or a few pieces of paper, and still be a contributing perceptual scientist.

This expansiveness is one of the reasons I find this field of research so congenial. Every few years, I become jaded with a task, turn to another, and then later find myself returning to what had been suspended earlier. However inconstant my day to day activities, my professional life for many years has been single minded. My long term goal has been to understand the “input” aspect of human cognition. The urge to achieve this goal has been undiminished for what is now, to my horror, almost a third of a century of a more or less constant commitment to sensory and perceptual studies.

In spite of the long term commitment on the part of so many in the past and present, it is clear that there are many unanswered questions concerning form perception. In fact, most of the relevant and significant questions so far formulated are as yet unanswered. We have hardly begun even to define the vocabulary of the science, much less understand how things happen.

Even more startling (than to appreciate that most relevant questions are still unanswered) is to realize in light of the long history I shall briefly review later in this chapter, how infrequently form perceptionists of the past or present have asked what is perhaps the fundamental question. That question, whose neglect a number of our contemporaries (e.g., Sutherland, 1967; Zusne, 1970) have also noted, is—What are the attributes or characteristics of a form that regulate its detectability or recognizability? It is important to appreciate that I use the word attribute here to emphasize that I am especially concerned with the global properties of the form rather the local features. I intend there to be a major difference between my use of the term “attribute” and the use of the word “feature.” It is the difference between some aspect of the overall arrangement of, as opposed to the nature of, the component parts of a form that I believe dominates form perception. In short, my thesis in these lectures is exactly that people see forms as a result of the arrangement of, and not the nature of, the component parts!

Since the heyday of the Gestalt tradition, relatively few psychologists have approached the study of visual form perception with such a global emphasis. Among these few are Rock (1973), Brown and Owen (1967), and Garner (1974). Even these globalists have usually emphasized some simple transformation (e.g., orientation), some general feature (e.g., compactness), or the influence of memory (and/or how memory or imagery are influenced by form) rather than the impact of the organizational geometry of the stimulus form itself on detection, discrimination, or classification. Indeed, recent decades have clearly been dominated by a strong elementalist tradition in perceptual theory. At first, spatial domain feature detection concepts dominated theory; currently frequency domain spatial frequency ideas provide the basis for what is clearly a consensus model. Both approaches are alike in stressing the importance of the component part, as opposed to global organization, in the determination of the perceptual response.

It is only in the last few years that the wholistic tradition has been notably revitalized, as evidenced by Kubovy and Pomerantz's (1981) recent book. Even in this extraordinarily thoughtful compendium of papers on perceptual organization—a book with which I felt an enormous amount of sympathetic agreement, and which substantially conforms to the approach, if not the experimental detail, of these lectures—there is relatively little consideration given to the global stimulus attributes involved in form perception. In some cases this is specifically because the authors reject the attribute approach, but in others, the empirical data is simply not obtained in a manner that might help develop an answer to this question.

One persistent and pervasive approach to “data” and “proof” in Kubovy and Pomerantz's book, as well as in classic Gestalt psychology, is the use of compelling demonstrations, rather than parametric experimental manipulation—the approach that characterizes so much of the rest of experimental psychology. Only in the articles written by Julesz, Pomerantz, and Shepard do we see the kind of parametric manipulation that seems to me necessary for understanding form perception. Nevertheless, the neo-Gestaltism reflected in this book is a promising sign that the wholistic tradition's contributions have not been lost and are regaining the attention of contemporary psychologists.

There are three practical reasons, beyond the decline of wholism, for the long neglect of the attribute question. First, there is as yet no adequate means of quantifying what we mean by the word “form;” thus it is difficult to precisely specify the attributes of a form. While some authors have suggested statistical families of forms that are alike in some general way, there is no single dimension along which form may be continuously varied comparable to electromagnetic frequency in color research or acoustic frequency in pitch research. Furthermore, neither the algebra of form proposed by Leeuwenberg (1969, 1971) nor the statistical algorithms for generating individual samples of broad classes of form (Attneave & Arnoult, 1956; Fitts & Leonard, 1957) have yet proved satisfactory and acceptable means of manipulating form as an experimental variable in the manner that scientific psychology depends upon so much. Such formularizations as Leeuwenberg's may model the psychological propensity to classify forms according to those general properties that are common to a group of forms (as has been pointed out to me in a personal communication from H.F.J.M. Buffart in 1982); nevertheless they do not define form in a specific enough manner to allow us to use these classifications as measures of an independent variable. In a sense all of these form generating methods are prototheories of how the visual system works rather than a practical means of scaling physical stimuli along a continuous dimension.

Perhaps the fundamental source of this difficulty lies in the fact that spatial and temporal forms are intrinsically multidimensional, and in present day psychology we still tend to think mainly in unidimensional terms. Forms, in the absence of a unique descriptive dimension are often generated in a more or less arbitrary manner and are equally often defined as experimental stimuli on the basis of some vaguely articulated ad hoc rule. This difficulty remains; my group has done no better than our predecessors in resolving the problem. As reported later, the stimulus forms we use are also more or less arbitrary, although in some cases a continuous variable (e.g., variance) does satisfy the immediate needs of a particular experiment. I must also acknowledge that it is entirely possible that the search for a precise quantification of “form” may be a search for a chimera; it may never be possible to quantify forms. This issue is yet to be resolved.

The second reason that the specific attribute problem has been ignored is that heretofore there has been no easy way to easily manipulate even arbitrarily defined forms in stimulus displays. Gestalt psychology was damaged perhaps even more by this practical difficulty than the falsification of their neuroelectrical field theories. Those pioneers simply did not have the technical capabilities to carry out the obvious experiments. One must wonder what the state of contemporary psychology would have been if those insightful psychologists had possessed the information manipulation tools now available to modern perceptual researchers. The advent of the laboratory computer, in particular, has ameliorated this practical difficulty. Forms of great variety and complexity in two, three, and even four dimensions (i.e., X, Y, Z, t) are today easily generated in many laboratories about the world.

I believe the third reason the specific attribute problem has been neglected is that the manipulation of the form of continuous figures usually leads to a confounded outcome. That is, changing one attribute of the global arrangement of the parts of a form also often covaries some other local feature. For example, varying the area of a geometrical form also varies the perimeter of that form. Such a confounding often makes the actual causal relationship between any particular attribute of the form and any measure of the perceptual response uncertain. The use of dotted stimuli sometimes overcomes this problem. There are, in the case of such stimulus materials, no local attributes other than “arrangement” itself; as long as the number of dots remains constant, all of the other aspects of the stimulus can be subsumed under the single factor called “arrangement.” On the other hand, “arrangement,” however singular it is as an item in our vocabulary, is itself not a simple term; it is at least as complicated as “form” and arrangements themselves may fall victim to multidimensional confounding in cases in which the intention is to change only a single attribute.

Nevertheless, dot patterns can often be manipulated in a reasonably straight-forward manner compared to continuous visual stimuli. For example, a line of five dots may be elongated from one to two centimeters without changing the number of dots in the line. The physical stimulus intensity is thus kept constant. A continuous line, however, can be elongated only by adding luminous area (the number of pixels along the line), and thus a long straight line produces more total physical energy than does a short line if the elements are kept equally luminous. The stimuli used in the experiments described in Chapter III serve as examples of the extent to which we have overcome this third difficulty.

It is my goal in these lectures to assay the ways in which the visual system responds to a set of arbitrarily designed and highly constrained dotted stimulus forms. It is my hope that by manipulating some of the attributes of these abstract approximations to continuous scenes and measuring their effects on form detection, that a few steps towards a general understanding of the ways in which we perceive geometric forms will be forthcoming. Obviously, this is an ambitious goal and not one that is likely to be fulfilled in the short run. Therefore, these lectures can deal only with a few experiments from which I shall attempt to draw some germane, but highly limited, conclusions as well as partially test one formal model.

Prior to a discussion of conclusions, however, I must direct your attention to some less earth-shaking, but practically important, details of method and some raw experimental results that may seem esoteric and isolated from the grand epistemological question (How do we see forms?). I hope my audience will not despair, because an understanding of these abstract experimental stimuli, technical methods, and empirical results is essential for a scientific (as opposed to an intuitive) solution to the problem of form perception. The absence of such concrete anchors to psychobiological reality would permit us to fall victim to the ruminations of armchair speculation. And, as we shall see, speculation without empirical testing in at least a few instances would have led us wildly astray; some of the results that are obtained in this study are surprisingly counterintuitive.

The specific attribute question is, however, only one of two major epistemological issues toward which our work is aimed. The other is the grand old question of perceptual research concerned with visual space—how is it that we are able to construct the third dimension from the two dimensional images projected on the retinae? That this is both a long standing issue and a perplexing one has been most eloquently expressed by my good friend and colleague Dan Robinson (1982) of Georgetown University when he states:

The very phenomenon of “space” erected so durable a barrier against radical empiricism that Mill, Bain, Helmholtz, and Wundt—these otherwise legendary empiricists—quailed before it. Helmholtz could explain space perception only by invoking the delphic process of “unconscious inference” and. in a similar vein, Wundt had to rely on something called “synthesis.” The details of their respective accounts are less important than their theoretical justifications: Since space is not given by any property of a stimulus, it must be constructed (inferred; synthesized) by the nonsensory (intellectual-cognitive) processes of the percipient. What this requires of Spencer and the Mills, “intoxicated with the principle of association” (James—Principles, 2:270), is the impossible task of accounting for the sensation of space through a compounding of totally nonspatial sensations. They, with Wundt and Helmholtz, may be lumped as “psychic stimulists,” ensnared by the very Kantian principles they so eagerly disown, [pp. 199]

I doubt that the vocabulary with which I am most comfortable (e.g., “perceptual space is calculated by the extraction of invariances existing in the two monocular images”) is likely to satisfy Dan Robinson any more than the vocabulary of “unconscious inference” or “synthesis.” It should not be overlooked, however, that in some functional way, these earlier theorists may have, in fact, been right! Even though they lacked the modern information processing metaphor necessary to phrase their theories in acceptable contemporary terms, they may have described the process appropriately in the less formal terms of their times. Space is implicit in the relationships between the two retinal images even if not “given” directly by any single attribute of the stimulus. Space is “given” indirectly in that it must be made explicit by more or less straightforward computational or transformational processes carried out on multiple aspects of the stimulus. Indeed, as we shall see, the stark reality may be that nothing is “given” directly, but rather that the superficial isomorphic relationship of the two dimensional experiences to the physical X and Y axis is no more “direct” than the relationship of the perceived Z axis to the dichoptic invariances. This tenative conclusion cuts through the knot posed by the question of how we see depth when the inputs are only two dimensional by asserting that everything is indirect (i.e., all perceptions are mediated by implicit representational mechanisms). Such an approach releases us from the obligation to find any special manner by which the mysterious Z-axis becomes like the apparently unmysterious X and Y axis by suggesting that all three are equally mysterious! One line of evidence would support such a seemingly far fetched and initially repugnant suggestion. If observers performed as well vis a vis the X, Y and Z axis...