July 1976
Memo on the Process of Perception: Invariance Detection
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 References to these essays must cite them explicitly as unpublished manuscripts.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 process of perception is one of information “pickup,” not one of information “processing.” Information is to be found in the flux of stimulation, not in “stimuli.” Information about the persisting environment, I have argued (IIA), lies in the invariants of structure. While information about events in the environment lies in the changes of structure. of the flowing stimulus array (IIC). The world persists in some respects and changes in some respects. Both the change and the non-change in the flux are detected by an active perceptual system.
The detecting of invariants is evidently at the heart of the problem of perception. But there is surely more than one level at which this occurs. Most fundamental is the perception of persisting surface layouts, objects, and entities together with the perception of non-persisting entities. How do we see a thing as continuing to exist and having existed as distinguished from a thing that is transient? (Old hints of this question are found in the puzzles of “object permanence” and “phenomenal identity”.) At quite another level is the perception of the similarities and differences among persisting entities that allow them to fall into classes or categories. How do we distinguish among things? Both of these levels of perception can be thought to involve invariant detection but the first case is concerned with invariants over time to specify a continuing entity and the second with invariants over entities. That which remains unchanged despite change is not that which remains unchanged despite difference. What are some examples of these different levels of perception?
I. Persistence and Change
1. Perceiving the constant layout of rigid surfaces and objects despite change of the point of observation. The perceiver sees an unchanging environment with only a change in his position. Beneath the changing perspective structure of the array there are invariants of structure to specify the layout. Any change of perspective structure is reversible, like the locomotion itself. Orthodox theory only recognizes these facts under the heading of shape and size constancy, referring to the plane faces of objects, their “forms”, that are projected in the optic array. Actually what one perceives is the layout or arrangement of surfaces, the combination of shape and size together with curvature, and the faces of objects on their far sides as well as their near sides. Nevertheless, some of the invariants of structure that specify layout are suggested by the invariants of classical projective geometry and the perspective geometry of “forms.”
2. Perceiving the persisting layout of those surfaces of the environment that interchange between hidden and unhidden. The hidden surfaces are always connected to the unhidden ones by occluding edges and these edges are specified by invariants in the optic array. Occluding is always reversible. But note that the shift, although lawful, is not a transformation in the mathematical sense of the term. This is a reformation of what is called the problem of the “phenomenal permanence of objects” in orthodox theory, the solution being based on concepts, memories, and experiences.
3. Perceiving the persisting identity of objects that move relative to the terrain. When a detached object is moved or turned it is seen as the same thing in a different position. All rigid Newtonian motions are reversible and the corresponding changes in the optic array are also reversible. Note that objects move but that places and the ground itself do not. The invariants to specify the persisting object have been described earlier. This fact of perception has been partially recognized in the old puzzle of the “transposable gestalt” which yields the same percept wherever it may fall on the retina, and in the puzzle of the “phenomenal identity” of objects.
4. Perceiving the persisting identity of an object or surface that changes shape. This refers to the surface of a non-rigid substance. A mechanical event changes its layout but does not rupture or break up the surface, certain topological invariants will persist. The change may or may not be reversible (depending on the elasticity of the substance). If these topological invariants are picked up the persistence of the object will be seen despite the deformation.
5. Perceiving the persisting identity of an organism or an animal. Organisms increase in size, that is, grow, and also decay. These changes are not reversible. Despite growth there are invariants of shape (D’arcy Thompson On Growth and Form). Such change is slow. A more rapid and more significant change is the deformation that goes with animal locomotion and behavior. Call it movement as distinguished from motion. Animate movements are usually (I think) reversible. The persisting identity of the non-rigid object we call an animal is perceived despite radical deformations of its surface. Note that they are often spontaneous movements not caused by an external application of force, the movements of altered posture, of locomotion, of expression, of communication, gestures, and of manipulation. The movements are perceived, especially those that specify a change in what the animal affords, and the persistence of the individual is perceived at the same time. (Only thus could the class into which the animal falls be perceived, but that is another matter). Only after all these preliminaries are we prepared to understand the vast experimental literature on the perception of the emotions of a person by “facial expressions” and the identification of an individual by the anatomical invariants of his face, that is, the “features” that distinguish it from other faces.
II. The Perception of Non-Persistence
The complement of the perceiving of the persistence of something is, of course, the perceiving of its non-persistence, that is, the ending or beginning of it. Surfaces begin with events such as crystallization, assembly from parts, and growth by self-division, and end with events such as dissolution, destruction, and death. These changes are clearly not reversible (Gibson and Kaushall film and Chapter IIC). The most radical change of a persisting entity is cessation, the ultimate change after which there are no invariants to be detected. I have argued that optical information is normally available to specify the going out of sight of a surface (by occlusion, by darkness, or by going into the distance) as distinguished from its going out of existence. To see a thing go out of sight is not confused with seeing it go out of existence except in the case when a conjurer, a magician, eliminates the optical information for its going out of sight and provides information that makes it seem to “vanish.” The discrimination may sometimes be difficult, however, and the child may have to learn it.
III. Types and Classes of Entities
We now come to another level of invariant detection, the perception of the sameness or difference of persisting things. Objects, places, organisms, animals, and artifacts in the environment are distinguished, at first vaguely and then more specifically, if they last long enough. They have both common features and distinguishing features, and they can be “compared.” They are the “same” or “similar” in some respects and different in others. The problem is familiar but actually ill-understood. We say that one pea in a pod is the “same” as another but also that they are not the “same” pea. Two letters in a text are the “same” letter but actually “different.” Logicians have worked on this puzzle as it applies to the names of things but not to the perception of things. The kind of perception now being considered is related to the classical problems of so-called concept formation. The detecting of same and different properties among entities is not the detecting of same and different properties of an entity that changes in time. The pickup of common features is not the pickup of persisting features. An invariant across objects is not the invariant of a persisting object. We pick up invariants over time in one way and invariants between objects in another way.
I suggest that invariant detection for a persisting object, place, animal, or person is simpler than invariant detection for a set of objects, places, animals or persons. Surely no act of comparison is required in the former case. When you continue to look at a thing you do not have to say “same as before,” and even when it moves or changes shape you do not say “same as before in this or that respect.” In this case I suggest that the perceptual system simply resonates to the invariants in the stimulus flux. It is attuned to the continuing structure. In the case of invariant detection for substantially distinct objects, however, the invariant may have to be extracted or even, for complex objects, abstracted from the variations.