Cognitive Psychology Issues for This Lecture · Size Constancy Size constancy is the perception than an object maintains the same size despite changes in the size of the proximal

1

Cognitive Psychology

Xiaolin Zhou

Department of PsychologyPeking University

Beijing, [email protected]

Issues for This LectureWe present examples of the distortion of perception; We discuss some of the ways in which we perceive a stable set of percepts despite constant changes in the size and shape of what we observeWe consider some of the perceptual information that leads us to perceive three-dimensional space from two dimensional sensory informationWe discuss theoretical approaches to perception

V1LGN

Perception

Perception is a process that use our previous knowledge to gather and interpret the stimuli that our senses register

it combines aspects of both the outside world (the visual or auditory stimuli) and your own inner world (your previous knowledge).

That is, it combines both bottom-up and top-down processing

Perception

perception in different modalities: vision, audition, olfaction (smell), gestation (taste), touchThe perception is much more than simply the registering of sensory information.

2

Illusions in Visual Perception (1)

Sometimes we cannot perceive what does exist

Illusions in Visual Perception (2)

Sometimes we perceive things that does not exist

Hermann Grid Illusions in Visual Perception (3)

At other times, we perceive what cannot be there

3

Sometimes we have difficulties in deciding which is figure, which is ground

Visual Visual AgnosiaAgnosia

An inability to recognize visual objects that is neither a function of general intellectual loss nor a loss of basic sensory abilities

Apperceptive agnosiaAssociative agnosia

Visual Visual AgnosiaAgnosia

Apperceptive agnosiaPatients are unable to recognize simple shapes such as circles or triangles, or draw shapes that are shownhave problems with information processing relatively early in the visual system

Associative agnosiaPatients are able to recognize simple shapes and can successfully copy drawings of even complex figuresHowever, they are unable to recognize such objects (e.g., Ratcliff & Newcombe, 1982)Have intact early processing but have difficulties with pattern recognition, which occurs later in information processing

4

Benson & Greenberg (1969)A soldier who suffered brain damage due to accidental carbon monoxide poisoning. He

was able to discriminate light intensities and colors, and to tell in what direction an object was movingcould recognize objects through their feel, smell, or soundbut was unable to distinguish a picture of a circle from a square or recognize faces or letters

His system was able to register visual information, but somehow his brain damage resulted in a loss of the ability to combine visual information into perceptual experience

What is perception for?

Registering the information that arrives at our eyes and earsPlacing some sort of interpretation on that information

Sensation, Perception, and Cognition

It is better to view these processes as part of a continuum, in which information flows through the system, with different processes designed to address different questionsQuestions of sensation focus on qualities of stimulationQuestions of perception are focus more on identity, form, pattern, and movementCognition occurs as perceptual information is used to serve further goal

Some Basic ConceptsDistal (external) objectInformation medium

reflected light, sound waves, chemical molecules, or tactile information coming from the environment

Proximal (near) stimulation (sensory registration)

when the information comes into contact with the appropriate sensory receptors

Internal perceptual objects (percepts)it in some way reflects the properties of the external world

The Variation of Proximal Stimulation

We can never experience exactly the same set of stimulus properties we have experienced beforeHow do we achieve perceptual stability in the face of this utter instability at the level of sensory receptors?Sensory adaptation

saccades, stabilized images

Stimulus variation is an essential attribute for perception

Stabilized images on retina creates no percept

Perceptual Constancies

An important way in which the perceptual system deals with variability is by performing analyses regarding the objects in the perceptual fieldPerceptual constancy occurs when our perception of an object remains the same even when our proximal sensation of the distal object changesOur perceptual system has mechanismsthat adjust our perception of the proximal sensation changes

5

Size ConstancySize constancy is the perception than an object maintains the same size despite changes in the size of the proximal stimulus

Perception of a person

Our sensory and perceptual systems can be misled by the same information that usually helps us to achieve size constancy

The Ponzo illusionMuller-Lyer illsion

The Ponzo Illusion

It occurs because of the depth cue provided by the converging lines

Muller-Lyer Illusion

Reasons not fully understood; the diagonal lines may be implicit depth cues similar to the ones we see in our perception of the exterior and interior of a building

Relative Size Illusion

The size of the center circle relative to the surrounding circles affect perception of the center circle’s size

Fraser Illusion Tilt Illusion

6

Repulsion Effect in Tilt IllusionSeveral Parameters Influencing the Magnitude of the Tilt Illusion (Smith & Wenderoth, 1999, Vision Research, 39, 4113-4121)Angles

Repulsion effect: < 60°, 15°Attraction effect: 60~90°, 75°

Test stimuli durationRepulsion effectAttraction effect

Contrast

Luminance Illusion

Shape ConstancyWhile size constancy involves the perceived distance of an object from an observer, shape constancy involves the perceived distance of different parts of the object from the observer.

An object’s perceived shape remains the same despite changes in its orientation and hence in the shape of its retinal image

A doorWhy?

Depth PerceptionHow do you perceive three-dimensional space when the proximal stimuli on your retinas comprise only a two-dimensional projection of what you see?Depth cues are either monocular or binocularMonocular cues: texture gradients, relative size, interposition, linear perspective, aerial perspective, location in the picture plane, motion parallaxBinocular cues: binocular convergence, binocular disparity

7

Binocular Depth CuesImpossible Figures

One of the confusing aspects of impossible figures is that there is contradictory depth information in different sections of the picture

8

多稳态多稳态 ((multistablemultistable) ) 视觉现象与视觉现象与

知觉的生理机制知觉的生理机制

Binocular RivalryGestalt Approaches to Form Perception

Perception organizes objects in the visual array into coherent groupsStructuralist approach - decomposional

It cannot explain how the dynamic whole of the structure might different from the sum of its parts

Gestalt approach a direct reaction to structuralism

9

Law of PragnanzWe tend to perceive any given visual array in a way that most simply organizes the disparate elements into a stable and coherent from.Gestalt principles

figure-groundproximitysimilaritycontinuityclosuresymmetry

Figure or Ground?

Illustrations of Gestalt Principles Illustrations of Gestalt Principles

Illustrations of Gestalt Principles Palmer (1977)Palmer (1977)

10

Does perceptual grouping involve attention?

Theoretical Approaches to Perception

How do we connect what we perceive to what have stored in our minds?

Perception is not as simple as association

The task of pattern recognition is to identify what these objects in the visual world are.

Two Types of Approaches

Bottom-up approachestemplate theoriesprototype theoriesfeature theoriesstructural-description theories

Top-down Approaches

Template-matching TheoriesA retinal image of an object is faithfully transmitted to the brain and an attempt is made to compare it directly to various stored patterns

These patterns are called templatestemplatesWe store in our minds myriad sets of templates, which are highly detailed models for patterns we might potentially recognizeWe recognize a pattern by comparing it with our set of templates and then choosing the exact template that perfectly matches what we observe

Fingerprinting, electronic code

Template-Matching Models

How matching is conducted?

Serial?Parallel?

Preprocessing?

11

Template-matching Theories

Things can go wrong very easily with a template. A mismatch occurs when the image

falls on the wrong part of the retinais a wrong sizeis in a wrong orientationis nonstandard letter’s

unwieldy and rigidFails to explain many aspects of perception, such as a ambiguous letter (A, H) in CAT or THE

In humans, pattern recognition is very flexible

Prototype Theories

A prototype is the best-guess (abstract, idealized) example of a class of related objects or patterns, which integrates all of the most typical (most frequently observed) features of the form or pattern

highly representative of a patterndoes not need precise, identical match; minor variations are allowed

We seem to be able to form prototypes even when have never see an exemplar that exactly matches the prototypeBut what about an ill defined category (e.g., building)?

Feature Theories

A pattern consists of a set of specific features or attributes

Stimuli are thought of as combinations of elemental features (e.g., horizontal, vertical, or diagonal lines, and curves)

Perception begins with the extraction of the features from the presented visual stimuliWe match features of a pattern to features stored in memory, rather than to match a whole pattern to a template or a prototype

Feature TheoriesFeature model has a number of advantages over the template model

Since the features are simpler, it is easier to see how the system might try to correct for the kinds of difficulties caused by template modelsIt is possible to specify those relationship among the features that are most critical to the patternUse of features rather than larger patterns will reduce the number of templates needed; same features tend to occur in many patterns

Pandemonium ModelPandemonium ModelSelfridgeSelfridge, 1959, 1959

Image demonsretinal images

Feature demonsmatching retinal images to features

Cognitive demonsfeature combinations

Decision demons

Pandemonium Model

12

Kinney et al., 1966

Letters are presented for very brief intervalsFor the letter G, 29 errors were made by subjects. Of these errors,

21 involved misclassification as C6 misclassifications as O1 misclassification as B1 misclassification as 9

Gibson, 1969

Same or different?

P R > G MThe number of distinctive features shared by the letters (Garner, 1979)

Perceptual ConfusabilityHierarchical Cluster Analysis

Local and Global Features

Global features: the features that give a form its overall shapeLocal features: constitute the small scale or detailed aspects of a given from

Local and Global Features

Global precedence effect (D. Navon)Perceptual processes proceed from global structure towards more fine-grained analysesHowever, it has not proved possible to identify precisely whether in the visual system the global advantage occurs

Local precedence effect (M. Martin)Global advantage occurs only when the global structure of a pattern can be ascertained by a single eye fixation

13

Global precedence effect Local precedence effect

Physiological and Neurological Support for Feature Theories

Hubel, 1982; Hubel and Wiesel, 1965, 1979Most of the cells in visual cortex respond only to specifically oriented line segments

feature detectors, receptive field

These cells seem to show a hierarchical structure in the degree of complexity of the stimuli to which they respondBut once discrete features have been analyzed, how are they integrated into a form we can recognize as a particular object?

“What” and “Where” Pathways

Separate neural pathways in the cerebral cortex for processing different aspects of the same stimuli

V1LGN

“What” and “Where” Pathways

The “what” pathway descends from the primary visual cortex in the occipital lobe towards the termporal lobes, and is mainly responsible for processing the color, shape, and identity of visual stimuliThe “where” pathway ascends from the occipital lobe towards the parietal lobe, and is responsible for processing location, and motion information.Thus feature information feeds into at least two different systems for identifying objects and events in the environment

14

Structural-Description Theory

A pattern is more than the sum of its parts

Relations among pattern features are importantDestroying relational information is

particularly detrimental for object recognition (Biederman, 1985)

Irving Biederman: recognition-by-component theory

Structural Theory

We form stable three-dimensional mental representations of objects by manipulating a few simple geometric shapes, i.e., a set of three-dimensional geons (for geometrical ions)Accordingly, we quickly recognize objects by observing the edges of objects and then decomposing them into geons, which also can be recomposed into alternative arrangements

BiedermanBiederman et al., 1985et al., 1985

BiedermanBiederman et al., 1985et al., 1985 Cave & Cave & KosslynKosslyn, 1993, 1993

15

Top-down Approaches

Constructive (intelligent) perceptionthe perceiver builds (constructs) a cognitive understanding (perception) of a stimulus, using sensory information as the foundation for the structure, but also using other sources of information to build the perception

Our perceptions of size, shape, and color constancy indicate that high-level constructive processes are at work during perception

Constructive Perception

During perception, we quickly form and test various hypotheses regarding percepts, based on

what we sense (the sensory data), what we know (knowledge stored in memory), and what we can infer (using high-level cognitive processes).

We perform unconscious inference to assimilate information from different sources to create a perceptionIntelligence is an integral part of our perceptual processing

Examples of Context Effects

This view is particularly good at explaining context effectsConfigural-superiority effectObject-superiority effect

Configural-Superiority Effect

Context and the Recognition of Context and the Recognition of Scenes (Scenes (BiedermanBiederman et al., 1973)et al., 1973)

Context and the Recognition of Context and the Recognition of Scenes (Scenes (BiedermanBiederman et al., 1973)et al., 1973)

16

Synthesizing the Two Approaches

Instead of viewing these theoretical approaches as incompatible, we may gain deeper insight into perception by considering the approaches to be complementary

sensory information may be more richly informative and less ambiguous than the constructivists would suggestperceptual processes may be more complex than hypothesized by Gibsonian theorists, particularly when stimuli appear only briefly or are degraded

A Computational Theory of Perception

David Marr (1982) proposed a theory that considers the richness of the sensory information without dismissing altogether the value of prior knowledge and experience in perception

Marr proposed that raw sensory data from the retinas of the eyes can be organized through the use of three kinds of features: edges, contours, and regions of similarity

Organizing Features

Edges form the boundaries between and around objects and parts of objects

Contour features differentiate one kind of surface from another other

Regions of similarity are areas that are largely undifferentiated by distinctive features

Marr’ Model (1)

A series of representations (i.e., descriptions) provide increasingly detailed information about the visual environment:The human brain uses a three-step process for computing a 3-D percept of what we see.

Marr’ Model (2)

First, the brain creates a 2-D primal sketch of the sensory information that reaches the eyes

this sketch takes sensory data regarding changes in light intensity and maps out edges, contours, and regions of similarity, entirely based on the observer’s sensations from a given viewpointThis sketch represents an object, e.g., a chair, in just two dimensions.

Marr’ Model (3)

Then the brain creates 21/2-D sketch of the data, which considers depth cues and surface orientations

the brain converts the primal sketch into the 21/2-D sketch, which enhances the 2-D primal sketch by considering the observer’s view of the orientation of surface, as well as depth cuesFor example, now the perception of chair includes some aspects of depth, showing the orientation (e.g., backwards) of the chair in the picture plane.

17

Marr’ Model (4)

Finally the brain creates a 3-D model, which represents three-dimensional objects and the spatial interrelationships among objects

this 3-D model includes the observer‘s spatial relationship with the objects but is independent of the observer’s viewpointObjects that were obscured from view would still be represented in the 3-D modelin forming the 3-D model, the prior knowledge and experiences may influence perception, although Marr did not specify how this influence might be excerted

Parallel Distributed Processing

Neural modelAll of the information available to the sensory field is processed simultaneously

All the components, including “parts” and “ wholes”

Knowledge resides in the connections, not the units connectedConnections could be excitatory or inhibitory, and could have different weightsConnection weights could be modified by input

Interactive-action model for word superiority effect

Questions for this Lecture

How do we perceive stable objects in the environment, given variable stimulation?What are the two fundamental approaches to explaining perception?

18

Reading List (1)

P. S. Churchland & P. M. Churchland. Neural worlds and real worlds. Nature Reviews Neuroscience, 2002, NovemberShinsuke Shimojo, Michael Paradiso, and Ichiro FujitaWhat visual perception tells us about mind and brainPNAS 2001 98: 12340-12341Binocular visual surface perceptionPNAS 1996; 93: 634-639.

Reading List (2)

Figure-ground activity in primary visual cortex is suppressed by anesthesiaPNAS 1998; 95: 3263-3268. D Purves and LE White

Monocular Preferences in Binocular ViewingPNAS 1994; 91: 8339-8342.

Reading List (3)

Is perception discrete or continuous? Trends in Cognitive Sciences Volume: 7, Issue: 5, May, 2003, pp. 207-213 VanRullen, Rufin; Koch, ChristofKeeping perception accurate Trends in Cognitive Sciences Volume: 3, Issue: 1, January 1, 1999, pp. 4-11 Bedford, Felice L.Visual illusions classified Trends in Cognitive Sciences Volume: 1, Issue: 5, August, 1997, pp. 190-194 Gregory, Richard L.Streams and consciousness: visual awareness and the brain Trends in Cognitive Sciences Volume: 2, Issue: 1, January, 1998, pp. 25-30 Milner, A. David

Reading List (4)

Constructing and representing visual objects Trends in Cognitive Sciences Volume: 1, Issue: 3, June, 1997, pp. 98-102 Singh, Manish; Hoffman, Donald D.More thoughts on perceiving and grasping the Müller–Lyer illusionNeuropsychologia Volume: 37, Issue: 13, December, 1999, pp. 1437-1444 Otto-de Haart, E. Grace; Carey, David P.; Milne, Alan B.