8/12/2019 Visual Illusions Classified

1/5

Gregory - Visual illusions

Visual illusions classif iedRichard L. Gregory

@@ @ objects and the working rules for seeinis necessary for any science because it facilit

a of illusions are no exception. It is hoped that the tentativean presented here may stimulate further research and raise n

ptual structure of how percepti. ,I ~e~~~~aIu~~~~~~,, ,;, _1

F llowing Helmholtz s lead we may say that knowledge isessential for vision because retinal images are inherently am-biguous (for example, the size, shape and distance of ob-jects) and because many vitally important object character-istics are not represented by images in the eyes (such ashardness, weight, hot or cold, edible or poisonous). ForHelmholtz, ambiguit ies are usually resolved and non-visualobject properties inferred, from knowledge or assumptionsby unconscious inductive inference. It is a small step to saythat perceptions are hypothesesa predicting unsensed char-acteristics of objects and the immediate future5.

That vision is but indirectly related to objects andstimu li was appreciated by Helmholtzs as following fromthe Law of Specific Energies (perhaps better named the Lawof Specific Qualities), proposed by his teacher JohannesMu r, which says: any afferent nerve signals the samequality or sensation whatever stimulates it. Thus, of course,we see colours not only from light but also when the eyes aremechanically pressed or are stimulated electrically. We mayregard eyes and the other sense organs as being designed bynatural selection to convey different stimu lus characteristicsby the same neural signals, action po tentials, which arerouted to different brain regions for the different sensationsof colour, touch, sound and so on. Sensations such ascolours are generated by specialized brain modules (Ref.7). It was clear to Newton* that it is strictly incorrect to saythat light is coloured. Light elicits sensations of colours insuitable eyes and brains, as we now say, psychologically pro-jected into external space to be seen as surfaces of objects .Thus, perceptions are projections of brain-hypotheses,which may or may not match physical reality.

Perception requires assumptions, which when not ap-propriate can generate cognitive illusions. A striking exam-ple is the hollow face. The visual bias that favours seeing ahollow mask asa normal convex face may be accepted as evi-dence of a powerful top-down object knowledge proces?.This bias is so strong rhat it counters the competing mon-ocular depth cues, such as shading and shadows of theconsiderable stereo information signalling that the image istruly hollow (see Fig. 1).It is significant that this, and very many other illusions ,are experienced perceptually even though the observer knows

conceptually that they are illusory. This does nor show thatknowledge has no part to play in vision; rather it shows thatconceptual and perceptual knowledge are largely separate,with the visual system being almost stand-alone. This is notaltogether surprising , for perception must work extremelyfast to be useful for survival, while conceptual decisions maytake minutes or even years. Furthermore, perceptions are ofparticulars rather than the generalities of conceptions .

For visual illusions to reveal brain or mind processesthey must be explained, and preferably placed in explana-tory categories. I Iere it may be noted that for perception it-self, sensory data must, at least implicitly, be classified forinductive generalizations to be possible. They must be clas-sified explicitly for deductive inferences. Classification bythe visual system is implicit , allowing Helmholtz ian uncon-scious inferences; but for us to relate the phenomena to the-ory, we need explicit classes of phenomena to consider.There are two basically different kinds of illusory phenom-ena: those having a physical or physiological cause, and cog-nitive illusions due to misapplica tions of knowledge.

Visual illusions due to disturbance of light between ob-jects and the retina are different from those due to disturb-ances of neural signals, but bo th m ight be called physical.Extremely different from these are cognitive illusions due tomisapp lied knowledge or misapp lied rules. The hollow faceeffect seems to be a clear example of the power of misplacedknowledge about a particular class of objects, namely, faces.Organizing rules, such as the Gestalt Laws of closure,proximity, continuity and the common fate of move-ments of parts of an object (such as leaves on a tree blownby the wind) are very generally applied, although not alwaysappropriately, to objects or situations. When object knowl-edge or the organizing rules are inappropriate, illusion canresult even though there is no fault in the physiology. Sothese (cognitive) explanations are not physiological, al-though , of course, physiology is always involved in percep-tion. We know a great deal about bottom-up processing ofvisual signals (especially from physiological recording fromthe striate area of the brain), which may introduce physio-logical signal errors; but this will hardly be considered here.We may distinguish between: bottom-up signals, top-down specific objec t knowledge, and what we may call

Copyright Q 1997, Elsevier Science Ltd. All rights reserved. 13646613/97/S17.00 PII: Sl364-6613(97)01060-7Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997

8/12/2019 Visual Illusions Classified

2/5

side-ways general rules. The knowledge and rules for per-ception may be compared with semantics and syntax of Ian-guage. Perceptions are regarded as hypotheses of objects andsituations (richer than available sensory data and indirectlyrelated to the external world). Perceptions are also predic-tive, so that behaviour is appropriate to non-sensed char-acteristics of objects, and anticipates events, to reduce orabolish physiological reaction-time. These considerationsmay suggest something like the ins-and-outs of vision ofthe functional scheme shown in Fig. 2.

This is a functional rather than an anatomical scheme,but it must be consistent with brain neuroanatomy. It hasbeen confirmed that there is a greater richness of fibresdownwards from the cortex, especially to the lateral genicu-late nucleus (LGN), than upwards from the eyes2-6. In ad-dition, in the primate b rain modulation of cortical process-ing by attention can occur.

The separation of conceptual and perceptual knowledgeis based on behavioural and phenomenal evidence, for ex-ample: percep tion must work within a fraction of a secondto be useful but conceptual decisions may take minutes oreven years. Furthermore, perceptions are specific but concep-tions may be general and abstract. Indeed,we experience an illusion although weknow it is an illusion, and even why it oc-curs. This separation is quite disturbing, asperception and understanding can disagree.

Classes of illusionsI will attempt to classify the phenomenaof illusions, in terms of appearances andcauses (see Ref. 18 for a useful source ofiiIusions and their histories). Visual illu-sions can be classified into four classesbased on appearance, which can benamed quite naturally from errors oflanguage: ambiguities, distortions, para-doxes and fictions (see Fig. 3) . It is poss-ible that language grew from pre-humanperceptual classifications of objects andactions. This may explain why languagedeveloped suddenly in biological time.

The causes of many illusory phe-nomena have not yet been explained sat-isfactorily, but classification may suggestwhere to search for answers. We maysuppose four principle causes, the firsttwo lying broadly w ithin physics; the lasttwo are associated with knowledge, and soare cognitive. The first (physical) is the re-sult of an optical disturbance interveningbetween the object and the retina; the sec-ond (physical) is due ro disturbed physio-logical signals in the eyes or the brain; thethird (cognitive) is the application of mis-leading knowledge of objects; the fourth(cognitive) is the application of mislead-ing general rules. Applying the appear-ances (named from errors of language,see Fig. 3), we arrive at Table I.

Fig. 1 The hollow face. The left-hand face is normal, the right-hand face is hollow.Viewing the actual hol low mask with bo th eyes, it appears convex, until viewed from asclose as a metre or so. Here top-down knowledge of faces is pitted against bottom-up in-formation from the eyes.

Conceptual knowledge

( Rules 1 Side-way:

/et for taskObjects exploredI Top-down

Boltom-upfrom hle eyes

Fig. 2 ins-and-outs of vision. The hypothesis generator is fed bottom-up by sensory data, and top-down bystored knowledge. These are shown in two largely separate boxes: conceptual knowledge is mainly abstract andgeneral; perceptual knowledge of objects is specific, and is used for interpreting or reading sensory signals, mak-ing them useful real-time data for object perception. General rules (such as the Gestalt Laws of perceptual orga-nization, and rules of perspective) are introduce d as side-ways. Learning from the results of actions by feed-backis very important, although some object know ledge and rules are innate, inherited from ancestral disasters. Theremay be some selection of knowledge and rules for the current task. The prevailing perceptual hypothesis (per-ception) may affect initial signal processing , downwards. (The evidence for this is from flipping ambiguous fig-ures such as the Mach corne r). It is specula ted that qualia of consciousness may serve to flag the present, normallyto avoid confusion with memory and imagination .

Trends in Cognitive Sciences - Vol. 1. No. 5, August 1997

8/12/2019 Visual Illusions Classified

3/5

Gregory - Visual illusions

Ambiguities People like us Necker cube

Hes miles taller Ponzo figure -/\ -

Language errors Visual illusions

Distortions

Paradoxes Dark haired blonde Tri-barimpossible figure000

Fictions They live in a mirror Kanizsa square t')ca

Fig. 3 Phenomena of illusions and errors of language. Visual illusions can be classified into classes: ambiguities, distortions, para-doxes and fictions, which can be derived from errors of language

Status of the classesThese classes are assigned tentatively, on the basis that per-ception is the active generation of hypotheses (perceptionsbeing best-bet predictive hypotheses of the object world)that are indirectly related to objects. The eyes signals maybe erroneous, and they may be misinterpreted, giving themost general classes: physical and cognitive (roughly,hardware and software).

Illusory phenomena are assigned to particular categoriesaccording to various criteria. Explicit litmus tests are re-quired to test the classification scheme. There are many ex-amples of litmus tests, for example, the loss of distortionwith colour but no brightness contrast ind icates retinal sig -nal errors, for example the cafe-wall distortionzO. Loss ofdistortions with seen depth indicates a cognitive expla-nation in terms of size scaling, as for the Ponzo, Muller-Lyerand Poggendorff distortions**. Changes of object as seenwith changes of orientation suggest the importance of objectknowledge, as for the inversions of Scott K imZ5 and thehollow face. There are many other li tmus tests but unfor-tunately they are rarely, if ever, made explicit in the litera-ture of perception and illusion. This is an important areafor future research.

QuahaThe most mysterious features of perception are the qualia ofconsciousness: experiences of red, green, pain and so on.Perhaps the most striking yet least considered visual experi-ment is the following: look carefully at an object, or scene,for several seconds; then close the eyes and see the imme-diate memory. Surely, the vivid qualia of vision have disap-

peared. Reversing the situation, compare the memory withthe visual experience when the eyes are opened. There is ashocking difference. It seems that normally qualia onlyoccur in real-time perception.

We neither know how qualia are produced by brainprocesses nor what, if anything, they do. I would like tospeculate that qualia serve to flag the present mome nt andnormally prevent confusion with the remembered past, theanticipated future, or more generally, with imagina tion.The present moment must be clearly identified for behav-iour to be appropriate to the present situa tion, and this is es-sential for survival. As top-down knowledge derived fromthe past is perhaps the major contributor to human percep-tion, real-time sensory inputs must not be ignored whensignalling the present.

On the Helmholtzian view of perception taken here, vi-sion is given mainly by inferences from limited sensed dataand from (not always appropriate) object knowledge andorganizing rules. As pointed out by David MarP, the eyesreceive only an observer-centred view of the world; but forinteractive behaviour with objects , perception must takeinto account currently invisible aspects, and non-sensedproperties such as hardness and weight. Clearly these are in-ferred; but this is an extension of the immense role of infer-ence in all object perception. It seems that the qualia of con-sciousness are limited mainly to what is sensed locally inreal-time; this being but part of perception, which is farmore general and largely non-conscious.

Visual time-confusion can occur in rare individuals hav-ing exceptionally vivid memory and imagination. A strikingcase is Mr M, described by the Russian neuropsychologist

Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997

8/12/2019 Visual Illusions Classified

4/5

Table 1. Illusions classified(Physical) (Cognitive)

Classes Optics Signals Rules ObjectsAmbiguity Cataract, mist, fog, Retinal rivalry, Figure-ground (usually Necker cube,

(any loss of image aberrant stimulation set by Gestalt Laws of face-vase, Schroderquality) (e.g. pressure on eye) closure, etc.) staircase, Mach

corner, hollow face

Distortion Astigm atism, Fraser spiral, cafe wall, Ponzo, Mijller-Lyer, Size-weight (smallspoon-in-water, Pulfrich pendu lum Poggendorff, Hering, object feeling heavierlateral inhibition (stereo-depth) vertical-horizontal than larger object ofeffects (such as (?), moon-illusion (?) same weight)brightness and colour (perception-depth)contrast). . . . . . . . . . . ....................................... .......................................

Paradox Looking-glass (ones Rotating spiral after- Tri-bar, Penrose Magritte mirror (backimage in wrong effect (expanding or triangle, Eschers of head seenplace, self in two contract ing, yet Belvedere, reflected, instead ofplaces at once) staying same size) Waterfall, etc. face)

Fiction Rainbows, moire After-images, Machs Kanizsa triangle, Faces-in-the-fire,patterns bands, Benhams disk Schumann (illusory man-in-the-moon,

(colour) contours and galleons in clouds,surfaces) ink blots

The Necker cube is the best known example of visual ambiguity (see Ref. 9). The hollow face illustrates resolving of ambigu-ity according to probabilities. The cafe walllllusion (long distorted wedges, seen in a chess board-like pattern, with alternaterows displaced by half a cyc/eJ depends on luminance but not on co/our contra@. The rotating spiral gives expansion (orcontraction) without change of size. The Ponzo, the Muller-Lyer and other perspective distortions are attributed to misplacedconstancy scaling (Refs 21, 22). The M-bar, impossible triangle (Ref. 23J and many more impossible figures can ppear to beimpossible objects when viewed from a critical position . The size-weight illusion is when smaller objects fee/ heavier thanlarger objects of the same scale weight. The Magritte mirror is a painting of the back of a mans head with a mirror impossiblyreflecting the back of his head instead of his face; our knowledge of mirrors, and the assumption that it is a mirror in thepicture, makes this appear impossible. (Paradoxes and puzzles of looking-glass images are discussed in Ref. 24.)

Alexandre Luria, whose memory and imagination were sovivid that he did have these problems of confusing thememory, for example of traffic lights, with actually seeingthemZ7. A more common exception to the usual lack ofvivid qualia in memory is emotional memories, such asmemories of guilt producing blushing. Perhaps this isq a-rich because there are afferent inputs from the bodyevoked by memories of embarrassmenP. Further excep-tions are dreams and the hallucinations produced by schizo-phrenia or hallucinogenic drugs - qualia that appear whenthe system is not functioning normally and is isolated fromsensory inputs or is stimulated abnormally by drugs.

The speculation is that normally for higher (cognitive)species, and especially humans, real-time sensory signalsevoke vivid qualia of consciousness to avoid confusion withmemory and imagination (which use much the same brainregions) as an aid to survival into the Future, by flaggingthe dangerous present.

References1 Helmholtz, H. won (1866) Concerning the perceptions in general, in

Treatise on Physiological Optics (Vol. III, 3rd edn) (translated bySouthall. J.P.C.). Opt. Sot. Am. 1925, Section 26, reprinted 1962, Dover

2 Gregory, R.L. (1968) Perceptual illusions and brain models Proc. R. Sot.London Ser. 8 171, 179-296

3 Gregory. R.L. (1968) On how so little information controls so muchbehaviour, in Towards a Theoretical Biology2, (Waddington, C.H., ed.),

Edinburgh. Also in Gregory, R.L. (1974) Concepts and Mechanisms ofPerception. Duckworth

4 Gregory, R.L. (1980) Perceptions as hypotheses Philos. Trans. R. Sot.London Ser. 13290, 181-197

5 Nijhawan, R. (1997) Visual decomposition of colour through motionextrapolation Nature 386, 66-69

6 Boring, E.G. (1950) A History of Experimental Psychology (2nd edn),Appleton Century Crofts

7 Zeki, S. (1993) A Vision of the Brain. Blackwell8 Newton, I. (1704) Opticks, Dover, reprinted 19529 Gregory, R.L. (1970) The /nte//igent Eye, Weidenfeld and Nicolson

10 Wertheimer, M. (1923) Untersuchungen zur Lehre van Gestalt. IIPsycho/. Forsch 4. 301-350. Organization of perceptual forms(translation) in A Source Book of Gestalt Psychology (1938) (Ellis, W.D.,ed.), pp. 71-88, Routledge Kegan Paul

11 Hubel D. (1988) Eye, Brain, and Vision, Scientific American Library12 Desimone, R. and Ungerleider, L.G. (1989) Neural processing of visual

information in monkeys, in Handbook of Neuropsychology (Eoller. F.and Grafman. J., eds), pp. 267-299, Elsevier

13 Fellernan, D.J. and van Essen, D.C. (1991) Distributed hierarchicalprocessing in the primate cerebral cortex Cereb. Cortex 1.1-47

14 Rockland, K.5. and Pandya, D.N. (1979 ) Laminar origins andterminations of cortical connections of the occipital lobe in the rhesusmonkey Brain Res. 179, 3-20

15 Van Essen, D.C. (1985) Functional organization of primate visualcortex, in Cerebral Cortex (Vol. 3) (Peters, A. and Jones, E.G.. edr),pp. 259-329, Plenum

16 Sillito. A. (1995) Chemical soup: where and how drugs may influencevisual perception, in The Artful Eye (Gregory, R.L. et al., edr). pp.294-306. Oxford University Press

17 Wurtz, R.H.B.. Richmond, B.J. and Newrome, W.T. (1984 ) Modulation

Trends in Cognitive Sciences - Vol. 1. No. 5, August 1997

8/12/2019 Visual Illusions Classified

5/5

of cortical visual processing by attention. perception, and movement, inDynamic A spect s of Neocortic al Function (Edelman, G.M., Gall, W.E. andCowen. W.M., edr). pp. 195-217, Wiley

18 Robinson, 1.0. (1972) The Psychology of Visua/Illusion. Hutchinson19 Gregory, R.L. (1971) The grammar of vision Listener83, 242. Reprinted in

R.L. Gregory (1974) Concepts and Mechanisms of Perception, pp.622629, Duckworth

20 Gregory, R.L. and Heard. P. (1979) Border-locking and the CafC Wallillusion Perception 4, 203-220

21 Gregory, R.L. (1963) Distortion of visual space as inappropriate constancyscaling Nature 199, 678691

22 Gregory, R.L. and Harris, J.P. (1975) Illusion-destruction by appropriatescaling Perception 4, 203-220

23 Penrose, L.5. and Penrose, R. (1958) Impos sible objects: a special type ofvisual illusion Er. 1. Psycho/. 49, 31-33

24 Gregory. R.L. (1997) Mirrors in Mind, W.H. Freeman25 Kim, S. (1981) Inversions, McGraw Hill26 Maw, D. (1982) Vision, W.H. Freeman27 Luria, A.R. (1969) The Mind of a Mnemonist Penguin28 James, W. (1890) Principles of Psychology, Macmillan29 Porner. M.J. and Raichle, M.E. (1994) images of Mind, Scientific American

Library

The Mind-Brain Continuumedited by Rodolfo Llinas and Patricia S. Church land, MIT Press, 1996.fA2.50 (xi + 315 pages) ISBN 0 262 12198 0The academic mind is addicted to prob-lems and averse to solutions , since aproblem is a nice little earner but a so-lution requires the unfortunate solverto find a new justification for the con-tinuation of funding. The mind-bodyproblem has been, and continues tobe, one of the nicest little earners of alltime. We learn in the introduction thatthe title of the present volume refersnot to the idea that the mind and brainare parts of a continuum, like thecolours of the rainbow perhaps (whichwould be an odd notion) but rather tothe denial of: the traditional Cartesianposit of a radical discontinuity be-tween the mental and the neural(p. ix). In its place the editors suggestcautious ly that: Mental activity, it ap-pears, is brain activity - caution beinga natural side-effect of an aversion tosolutions . So the continuum in the titleis, according to this view, a continuumbetween two things that are actuallyone and the same thing. Again, an oddkind of continuum, and hardly an im-provement on the more familiar for-mulation of this position as one ofmind-brain identity.

However that may be, the con-tributors happily steer away from thedeep waters of the introduction andthere follows a haphaza rd mix of re-views in various areas of cognitiveneuroscience, none of the contributorsmaking any reference whatsoever tothe supposed theme put forward bythe editors. An elegan t and impressivechapter by J.A. Simmons on images inbat sonar is, unfortunately, not likelyto be comprehensible to any philoso-pher who wants to find out how itfeels to be a bat. The most excitingchapter is by Logothetis and Sheinbergand it deals with the visual recognitionof objects in rotated views. They con-clude, from a wide survey of psy-chophysical and electrophysiologicalfindings, that object constancy de-pends on a memory-based, viewer-cen-tred recognition system rather than onan object-cen tred representa tion. Thebook ends with a chapter by

Churchland that finally does addressthe issues raised in the introduction.The mind-brain identity position is de-fended in a vivid and informal style:volcanoes, atoms and ferns are justout of luck as far as consciousness isconcerned (p. 290). On page 295 themisprint celebrates its ability to outwitthe spell-check and the grammar-checkby castigating Dennett for his beliefthat: chimpanzees and oranges arenot conscious. Now that the uncon-sciousness of oranges has been publiclybrought into question the founding ofPeople for the Ethical Treatment ofFruit cannot long be delayed, to be fol-lowed by liberation movements forferns and volcanoes.

Clearly then, any reader who goesto this volume in search of a roundedpresentation of the mind-brain iden-tity position, including objections to it,will be disappointed. The idea thatmental activity is brain activity has re-tarded resea rch in neuroscience . Wehave gone into the brain expectin g tofind such things as memories and per-cepts waiting there to be discovered,and systems for attention, action, andso on -all corresponding to traditionalmental events or faculties. The betterwe understand any of the brainprocesses we study, however, theclearer it becomes that they do not cor-respond in any sense to mental activ-ities in folk psychology. According toDuncan, for example, the concept ofattention for a neurosc ientist is, atbest, an intuitive label for a set of ques-tions whose answers in the end aregiven by reference to a general ac-count of cortical competition, an ac-count that goes far beyond the topicof attention. The search for an atten-tion system or an attentional mecha-nism in the brain has hindered, nothelped the finding of these answers.Similarly, the search for memory sys-tems in the brain, and the idea thatcortical p lasticity will instantiate the as-sociationism of folk psychology, havehindered progress*. This hardly seemssurprising if one considers that the

claim to remember an event is not adescription of ones brain, but rather amove in a rule-gover ned social gamecarrying, for example, the impliedpromise that one can supply furtherdetails about the event if requested. SoI am sceptical about the idea, pre -sented in Churchlands chapter, thatneuroscience is simply an improvedversion of folk psychology and thatfolk psychology is nothing more thanneuroscien ce-in-waiting . Rather, thetwo are paralle l activities wi th differ-ent aims, a situation that one couldtendentiously describe as: the posit ofa radical discontinuity between themental and the neural but more prop-erly is described as common sense. Isuppose most practising neuroscien-tists would agree with me, and wouldnot consider the relation betweenmind and brain as problematic, but inconsideration of their colleagues ad-dictions and aversions they would for-bear to say such a thing.David GaffanDept Experimental Psychology,Oxford University,Oxford, UK OX1 3UDtel: i-44 1865 271444fax: +44 1865310447e-mail: gaffan@psy .ox.ac. uk

References1 Duncan, J. (1996) Cooperating brain systems

in selective perception and action, in:Attention and Performance XVI (Inui, T. andMcLelland, J.L., eds), pp. 549-578, MIT Press

2 Gaffan. D. (1996) Associative and perceptuallearning and the concept of memory systemsCognitive Brain Res. 5. 69-80

Trends in Cognitive Sciences - Vol. 1, No. 5, August 1997