March 1967
Optical Occlusion and Edge-Information in an Optic Array
J. J. Gibson, Cornell University
The World Wide Web distribution of James Gibson’s “Purple Perils” is for scholarly use with the understanding that Gibson did not intend them for publication. References to these essays must cite them explicitly as unpublished manuscripts. Copies may be circulated if this statement is included on each copy.
The task of “ecological” optics is to formulate the principles of structure in the ambient optic array from the environment (p. 187-208 in The Senses Considered as Perceptual Systems) and to try and establish the usual environmental causes of this structure (p. 208-220) so as to understand the information contained in the light.
An important problem is to distinguish between the stationary structure of ambient light, as exemplified in the perspectives of an ordinary pictorial array (ch. 11) and what might be called its kinetic structure. By this I mean what are vaguely called the “motions” in an optic array, arising from either the movements of the observer himself (motion parallax) or the motions of an object in the world. There is always more information (in the sense of less ambiguous information) in a kinetic array than in a static array. The possible kinds of “motion” in an optic array have not been fully described; only a start has been made (e.g. p. 201-208). This is an extension of the previous Cornell efforts to describe such “motions and transformations.” The stationary structure of ambient light is fairly familiar to us because we know about the “law of the visual angle,” and the “laws of perspective.” The kinetic structure, however, if the term be allowed, is much less familiar.
Consider the principles of optical occlusion, that is, the hiding, covering, superposition, or screening of one thing by another. The “sight-lines” in the cross-sectional diagram herewith illustrate these principles. We should consider both stationary occlusion for a static array and kinetic occlusion for an array containing “motions” (see figure). We should note, however, that the static array is an almost unrealized limiting case, since a human or animal observer is never frozen at a single station point for any length of time, and a completely frozen environment is a fixtion. What are the principles of static and kinetic occlusion?
1. Surface layouts that do and do not involve occlusion. Static occlusion. We note first that an environment consisting only of flat “ground” or “background” will not manifest any kind of occlusion. There will only be unbroken gradients of the density of optical texture, or tuxture perspective. The ambient light from a theoretically bare earth entiails merely decreasing angular size of the optical texture with increasing distance out to the horizon (Figure 9.2 in Senses Considered). The cloudless sky contains no texture and no gradients. Like the earth, an enclosed space such as a bare room with floor, walls, and ceiling but without furnishing entials merely these same gradients of texture along with abrupt changes of gradient where the changes of slant of the surfaces occur — the “corners” (Fig. 10.8 in Senses Considered).
However, the environmental layout always in fact includes things like “u>detachable objects, or apertures, or ells in a wall, or cliffs, or overhangs. In solid geometry, a detachqable object is a topologically “closed” surface; an aperture is a topologically closed “hole” through a surface or object. These departures from mere background – layout always involve an occluding edge. An edge is projected in an optic array.
Projected and Unprojected Surfaces in a Stationary Optic Array
The surfaces indicated by thick lines are projected, and those by thin lines are occluded (“hidden,” “screened,” “covered”) at this station point. There are seven occluding edges in this diagram, and four corners. The converging lines do not represent rays of light but lines of sight. When the station point moves, seven of these margins in the array will convey edge information, and the direction of edge-depth, by the “gain or loss” of optical texture. Compare this picture with Fig. 10.4 (p. 193) in The Senses Considered as Perceptual systems, where the transformations of perspective forms of surfaces is emphasized instead of the edge information. All occlusion is temporary occlusion.
The “forms” of Gestalt theory, and the figure-ground phenomena that ensue when a closed contour is presented in a visual field (“the ground seems to extend uninterrupted behind the figure”) are cases of static occlusion represented in a pictorially frozen array. The fact that a closed contour on a black field almost always suggests a detached object in front of a vague surface and almost always suggests an aperture or window in that surface implies only that. ecologically, the perception of detached objects is commoner than the detection of apertures, windows, or holes. The rule that the phenomenal contour “belongs to the figure and not to the ground” reflects a probable, not a certain, fact of optical occlusion, since windows in a large wall do sometimes exist in the world. And the so-called “cue” of superposition (or interposition) in outline drawings turns out to be, when analyzed experimentally, not as simple as the rules of Rubin suggest (e.g. Woodworth Expt’l Psych., p.630). Phenomenal superposition in drawings depends on properties of the continuity of the outline (e.g. Chapanis and McCleary, 1953, J. Gen. Psychol.).
The specification of occlusion in a frozen optic array is less than perfect. Ambiguity may arise in certain circumstances as to object or aperture, as to cliff or overhang. The further reduction of information in outline drawings leads to such ambiguities as “impossible” objects and other anomalies (e.g. Hochberg, Perception, p. 84; Senses Considered, p. 247-248.
2. The occluding edge. An occluding surface necessarily has an occluding edge which hides, covers, screens, or interposes. There seem to be two geometrical types of occluding edges, the dihedral angle and the curved surface. The sight line in the optic array is that which touches the vertex of the plane angle or that which is tangent to the curve (study by Shepela).
An occluding edge is usually but not necessarily projected as a contrast margin in the optic array (a discontinuity of intensity). It is usually but not necessarily projected as a jump in the density of optical texture (a discontinuity in the gradient of size perspective). It is usually but not necessarily projected as a discontinuity in the gradient of binocular disparity (not when vision is with one eye). But I suggest that it is necessarily projected as a kind of discontinuity in a kinetic array, one in which either the station point moves relative to the environment or the occluding edge moves.
Note that the occluded surface is always more distant than the occluding surface from the observer. The direction of this depth at the edge is implied by the fact of occlusion. For an object on a ground the depth increases outside the cloud contour; for a window in a wall, the depth increases inside the closed contour. For a cliff the depth increases above the contour; for an overhang the depth increases belowthe contour.
3. Projected and unprotected surfaces in a temporary optic array. An optic array is a projection by sight-lines to a convergence point in a medium of a surface layout. The convergence point, the “point of view,” may be temporarily stationary but is usually moving. Except for the “bare” layouts described above (and except for transparent surfaces, which are not here considered) part of the layout is always projected (not occluded) and the remainder of the layout is unprotected (occluded) as illustrated in the diagram. In short any ordinary environmental layout may be divided into two parts, the projected and the occluded, for any given stationary array at any one time. This theoretically includes the whole set of faces and facets of the layout, large or small, near or far. In ordinary language, we refer to occluded surfaces as the back of an object, the background behind it, the backside of a wall, the surface extending behind a window, the far side of a hill, the valley hidden by it, etc. It has long been a problem in the theory of perception and learning as to how men can be aware of occluded objects, and why animals behave as if they knew where hidden objects and place are. These are not projected as patches of color in the visual field. They seem to have no basis in sensations. How can they be apprehended?
4. The change from unprojected to projected surfaces. Ecological optics makes the postulate of connected sets of station points in the dense intersecting network of sight-lines that fills an illuminated medium. These sets are paths of potential locomotion. A corollary is the “broadcasting” of the perspectives of an object by lines of sight throughout the medium, that is, to a manifold of station points where a crowd of observers might stand or a series of station points which a single observer might occupy.
The principle is that, for a world where observers and objects are not fixed in place, all occlusion is temporary occlusion, i.e., all unprojected surfaces can become projected. It can be shown, for any surface in any layout of surfaces, that there is some station point to which it is projected (at which it is projectively visible).
5. Two types of transition from occluded to disoccluded. An unprojected surface may become projected in either of two ways: by a change of station point (locomotion) or, if it is part of a “detachable” object, a motion of the surface. The latter involves a displacement or rotation of an object. All motions, either of station points or right objects, are reversible. This reversibility of the connected sets has mathematical consequences.
6. The change in the optic array during progressive occlusion and disillusion. We now come to the heart of the question of edge-information and of how one thing is perceptible behind another. Recent experiments with kinetic displays at Cornell and at the Lincoln Laboratory, not yet published, seem to suggest a formula for what happens optically during progressive occlusion and disillusion. These experiments are somewhat related to Michotte’s work on the “tunnel effect ” but instead of using a form for the display they use a random texture. They are connected with earlier Cornell studies on the separation in depth of two apparent surfaces (Gibson, Gibson, Smith, and Flock, 1959) but instead of transparent depth they try to isolate edge-depth. The formula to be suggested is not the same as that proposed on p. 204-205 of the Senses Considered, based on Michotte.
The present hypothesis is that change of occlusion entails a unique kind of change of structure at the margin projected by the edge. On one side of the margin optical texture is “maintained” or “preserved” (corresponding to an occluding surface) while on the other side of the margin optical texture is either progressively lost (corresponding to a surface being revealed or uncovered). The side where the structure of the array is preserved is always in front; the side where the structure is gained or lost is always behind. Note that, on this formula, it makes no difference whether the occluding surface or the occluded surface is the one that absolutely moves. Hence the information for the edge and for the direction of depth does not depend on motion as such but on the gain or loss of structure on one side of a margin.
In experiments with random textures, the display is so arranged that whenever the progressive change ceases, the array becomes unbroken and continuous. The perception then reverts to a continuous surface with no edge and not even a visible margin, since no discontinuity of intensities, of density, or of disparity exists.
We may now consider an alternative to the above formula. Could the stimulus information be described as a progressive rupturing of the continuity of a textured array, without the feature of gain or loss of adjacent elements on one side of the margin? Such would be an abrupt shearing or “slicing” of the texture. This would occur when a rectilinear edge of an object moves over a background on a path exactly parallel to the edge, or when a belt moves behind a window exactly parallel to one edge of the window, or when an observer moves his head exactly parallel to the straight edge of a cliff, say. This pure “slicing” is a limiting case, therefore, which could not be realized except experimentally. (A moving circular object on a ground, or a circular window in a motion apparatus would yield this case only at two points on the circumference). If the pure case of a slicing motion without gain or loss were isolated experimentally, it might yield the perception of a separation of two surfaces, a margin, but not of an occluding edge with the property of covering and of being in front of. This prediction could be tested.
There are other alternatives to the above formula: a margin at which texture is gained on both sides, or lost on both sides, or gained on one side and lost on the other. The se cases have been produced in a display and recorded on film by G. Kaplan. The method was optically difficult and the screened motions are jerky. Observations are somewhat variable, but the tentative conclusion is that such cases do notyield univocal perceptions of an occluding edge with covering or superposition of one thing on another, whereas the margin at which texture is progressively incremented or decremented on one side only do yield perceptions of a surface being uncovered or covered, respectively, by an occluding edge. The anomalous and interesting perceptions obtained with the other cases are worthy of further study but they do not seem to be relevant to the problem of the information for the perception of an occluding edge as defined in this paper.
Michotte’s studies of the occlusion of a form (such as a disk or rectangle) at the “entrance” of a “tunnel” and its disocclusion at the “exit” (e.g. Michotte, Thines, and Crabbe, 1964) can perhaps be subsumed under the above formula if one states it as the progressive incrementing or decrementing of adjacent parts of a figure. The contour figure in an empty field is not, however, as typical of optical information as the Gestalt theorists supposed; a structured or textured figure is more common. The progressive adding or subtracting of parts of such a figure would fit the above formula. It should be noted that this is not well described by the ordinary meaning of the term “transformation.” The projected surfaces in an optic array may be described as undergoing transformation; what we are concerned with are the changes or transitions from projected to unprojected surfaces and the reverse (sections 4 and 5 above).
7. The optical information for a thing’s going out of existence. Occlusion, as described in this paper, is the ordinary way in which an object goes out of sight. It is usually hidden, covered, or screened from view before its projection becomes vanishingly small at the horizon of the earth. But an object may also go out of existence. This may happen, for example by decay, evaporation, dissolution, extinction, or other processes. The terms “disappear” and “vanish” have both of these meanings, according to the dictionary: to go out of sight and to go out of existence. These events are not distinguished verbally, but it is possible that they are distinguished visually by ordinary observers, and perhaps by animals and children. Michotte has shown that an object seems to persist phenomenally when it goes into a “tunnel” or (as may now be suggested) when it is optically occluded. Does an object not seem to persist phenomenally when it is not optically occluded but when its projection in the optic array undergoes some otherkind of change? If there is optical information for going out of sight, as our experiments strongly suggest, is there also optical information for going out of existence?
A beginning has been made in answering this question. Reynolds (1966, appendix to Cornell thesis) has presented observers with a moving luminous rectangle which can either fade into the background (by loss of luminance) or can be progressively wiped out at its leading contour by a dark rectangle. Observers are given a choice of reporting that the object “disappears and ceases to exist” or that it “disappears and continues to exist.” They almost always give the first report to the first display and the second report to the second display.
Other kinds of stimulus information than the fading of contrast are possible for the perception of a thing’s ceasing to be. They can be represented by the method of frame-by-frame photography, with animated drawings or paper-cutting, or perhaps with other methods. Such experiments are promising and the study of this phenomenon should be continued.
A film has been produced by animation entitled The Change from Visible to Invisible. A Study of Optical Transitions (Gibson, Reynolds, Wheeler and Kaplan, 1966). The shots, with titles, illustrate occlusion of one textured surface by another and occlusion of one textured face of an object by its adjacent face when the object is turned. Also illustrated is a circular figure that is “eaten up.” These figures do not have a textured background however; they are displayed in an empty field and this, I have argued, is not representative of the structured optic array from an ordinary environment. It would be interesting to see a figure that fades into its background, or disintegrates to merge with its background. But the existing film does provide some evidence for the conclusion that objects persist or endure in experience when the optical information specifies occlusion and do not do so when the information is of another sort.
The ancient problem of how animals and men become aware of a “permanent” environment outside the field of view is more comprehensible if we contrast it with the problem of how they detect the impermanence of things that physically change form one state of matter to another. Perhaps both kinds of apprehension depend on the information available in light.