Cognition Week 3

Perception

Sensation

detection of physical energy

Perception

organisation and interpretation of sensory information that produces our experience

Bottom-up processing

Perception begins with sensory information and builds towards meaningful representation

Top-down processing

Using prior knowledge, expectations and context to interpret sensory input

Gibson’s (1950) direct (ecological) theory of perception

The enviro contains sufficient info for perception The purpose of perception is to guide behaviour. Perception occurs directly from sensory information, without the need for extensive cognitive inference. Our perceptual systems have evolved to detect this information efficiently

Invariants - info available from the enviro

The sensory info changes as we move through the world, but certain relationships remain stable, allowing us to perceive objects and events directly

The visual system can directly detect stable patterns in the enviro, including:

Optic flow, Motion parallax, Texture gradients, Affordances

Optic flow patterns

Pattern of movement across the retina due to self-motion - What changes as we move? Provides information about the: direction of movement, speed, distance

Motion parallax

The relative speed difference between objects at varying distances, which creates the visual impression of depth

• Nearby objects move rapidly across your retina

• Distant objects move more slowly

• Very distant objects appear stationary

Texture gradients

Evenly spaced elements appear closely packed together as the distance from the viewer increases

Affordances

Action possibilities that objects or environments offer An affordance depends on both: the physical features of the object/enviro, and the physical abilities of the organism.

Exist in the enviro regardless of whether anyone perceives them

Gibson’s perspective on learning perception

suggested that we don’t need to learn to perceive

Visual cliff experiment – do infants perceive depth?

Table split in half (midpoint=ridge), one side shallow, other deep. Mother encourages baby to come to deep/shallow side. Babies more likely to to shallow side. Concluded: depth doesn’t require learning

Critiques of Gibson and Walk

Duration of infant crawling experience predicts avoidance of the visual cliff

• at 7.5 – 8.5 months infants: 35% of inexperienced crawlers avoid the visual cliff as compared with 65% of more experienced crawlers

• Kittens raised in the dark (27 days) do not avoid the visual cliff

Context predicts deep side approach/avoidance

• 12-month-old infants: if mothers smiled, infants more likely to crawl to deep side cliff, but if mothers made a frightened face, infants avoided the cliff

Top-down processing in perception

Knowledge/expectations the observer brings to the situation influence interpretation of sensory info.

e.g. what we expect to see in different contexts influences our interpretation of the identity of the “blobs” inside the circles

Phoneme restoration

Participants heard spoken sentences in which a word mid-sentence was partially covered by a cough. Participants had the perception of hearing the whole word

Perception is an active construction of reality

Both bottom-up and top-down processes interact to produce our perceptual experiences

Gestalt Principles

describe how we group visual elements into objects and patterns. Descriptive—they explain what tends to happen during perceptual organisation, but not the underlying cognitive or neural mechanisms. The visual system organises sensory input into coherent, meaningful wholes

Principle of Prägnanz (“Good Form”/ simplicity)

The visual system tends to organise sensory input into percepts that are simple, stable, regular, symmetrical and orderly

Figure ground segregation

where are the boundaries of objects

Proximity

objects near each other form a group

Similarity

Similar visual elements form a group

Continuity

Elements along a path are perceived as related

Law of closure

Objects are perceived as a whole even if they are not complete

Limitations of Gestalt Principles

Descriptive rather than explanatory

Limited account of top-down influences

Limited account of individual differences, experiences and expertise

When multiple Gestalt principles apply, it is unclear which principle will dominate or how the visual system resolves competing interpretations

Expectations from regularities in the world

Light comes from above - shadow underneath: perceive indentation. shadow above: perceive bump. Checkerboard illusion: A darker than B but they’re the same (shadow→amplify brightness)

Effects of context on object recognition

A. Appropriate for the context B. Inappropriate for the context but visually similar to A. C. Inappropriate for the context and visually dissimilar to A. Asked to write down what they saw; measured accuracy, good at saying A, 2nd best reporting C, worst B

Challenges for object perception: The visual system must recognise objects despite changes in:

Viewpoint – the same object looks different from different angles

Size (distance) – retinal image size changes as objects move closer or farther away

Lighting – shadows and illumination alter an object's appearance Occlusion – objects are often partially hidden by other objects

Background – objects appear in many different contexts

Within-category variation – no two examples of the same object are identical

Template matching models

Have a representation in mind of everything you see exact match or if wrong, try other templates - would take time, exhausting, assumes unlimited storage

Feature based models

Instead of matching whole objects, we extract features such as edges, corners, line orientations, etc. Only explains basic stimuliEvidence: neurosci: electrode in visual cortex of cat, can see diff activity in neurons depending on orientation . Vertical - heaps of activity, horizontal - no response

Recognition-by-components (RBC)

Once we’ve extracted basic features start processing in more complex ways. All objects are made up of 36 geons (geometric ions: basic elements make up complex objects)

Limitations of the RBC theory

Emphasises bottom-up processing, with relatively little role for topdown influences

Best suited to recognising basic object categories, but less successful at explaining fine-grained within-category discrimination

Assumes objects can be decomposed into a small set of invariant geons, but many objects are highly complex and variable

Does not fully explain the effects of expertise, where experience improves recognition of highly similar objects

Hierarchical visual processing

Feedforward processing: Information flows from lower to higher levels Each stage extracts increasingly complex information

Recurrent processing: Higher cognitive areas send information back to earlier visual areas