Perception and Sensory Neuropsychology

Colour is psychophysical, not physical

We see 400-700nm of the electromagnetic spectrum, mostly as reflected light

Photopic vision

Bright light; cones are used

Scotopic vision

Dim light; rods only, so no colour

Three cone types

S, M and L (short, medium, long wavelength); more accurate than "blue, green, red"

L-cone peak wavelength

About 565nm, closer to yellow than red

Univariance

One photoreceptor type can't tell wavelength from intensity, because different wavelength-intensity combinations give the same response

Why can't rods see colour?

All rods contain rhodopsin, so univariance applies

Trichromacy

Colour is defined by the outputs of the three cone types

Metamers

Different wavelength mixtures that look identical

Opponent colour theory: three channels

Red-green, blue-yellow and black-white

Cone-opponent cells in the LGN

Have centre-surround organisation

Hering's observation

"Reddish green" and "bluish yellow" can't exist

Negative afterimage

E.g. red leaves a green afterimage; shows opponent processing in action

Colour blindness prevalence

About 8% of males and 0.5% of females

Protanope

No L-cones

Deuteranope

No M-cones

Tritanope

No S-cones

Colour-anomalous

Two similar cone types

Cone monochromat

One cone type

Rod monochromat

No cones; badly impaired in bright light

Colour contrast

A region induces the opponent colour in its neighbour

Colour assimilation

Neighbouring colours bleed into each other

Colour constancy

Surfaces look the same colour under different lighting because we discount the illuminant; expectations play a role

RGB colour space

Based on cone outputs

HSB colour space

Hue, saturation, brightness

Synesthesia

One stimulus evokes another sensation (e.g. grapheme-colour); about 4-5% of people

What pathway (ventral stream)

Identifies objects, their names and functions; travels to the temporal lobe

Where pathway (dorsal stream)

Processes object location, movement and spatial information; travels to the parietal lobe

Split beyond V1

Visual information divides into two pathways for object recognition and spatial processing

Boundary ownership (V2)

Which side of an edge belongs to the object; V1 edge detectors can't do this

Figure-ground assignment

Automatic; smaller and more symmetrical regions tend to be seen as figure (e.g. vase/face illusion)

Mid-level vision

Sits between feature extraction and object recognition; handles texture segmentation and grouping

Gestalt principles

Good continuation, proximity, similarity, symmetry, parallelism

How does camouflage work?

By defeating the Gestalt principles

Navon's global superiority effect

We process the big picture (global) before the details (local)

Global shape processing

Relies on specialised whole-shape detectors

Radial frequency (RF) patterns

Used as evidence for global shape detectors

RF evidence 1

Sensitivity to RF shapes is better than V1 cells working independently would predict

RF evidence 2

Thresholds halve with each added part of an RF3 triangle, but adding parts to an RF24 barely helps

Attneave (1954)

The informative points on a contour are curves and angles

Hoffman & Richards (minima rule)

Convexities define object boundaries; concavities segment parts

Gallant (V4)

V4 processes curvature in macaque and human; damage to human V4 impairs curved shapes but not orientation or contrast

Pasupathy & Connor polar-based coding model

V4 neurons are tuned for curvature and the angular position of that curve relative to the shape's centre

Benefits of polar-based coding

Size invariance and an object-centred representation

Adaptation evidence for polar coding

Angular location matters, not physical distance between curves

Klüver & Bucy

Temporal lobe lesions in monkeys caused "psychic blindness", now called agnosia (seeing but not recognising)

Inferior temporal (IT) cortex cells

Huge receptive fields; selective for complex features (hands, faces, trained objects)

Grandmother cells

Idea arising from IT hierarchical models; single cells selective for one specific thing

Single-cell evidence in epilepsy patients

Cells selective for the Sydney Opera House, the Eiffel Tower and Jennifer Aniston

PPA

Parahippocampal place area: places

EBA

Extrastriate body area: bodies

FFA

Fusiform face area: faces

How were PPA, EBA and FFA found?

Subtraction method in fMRI; decoding fMRI patterns can predict what is being viewed

Why is object recognition hard?

Changes in viewpoint, size, lighting and shadow, yet we keep object constancy

What matters most for object recognition?

Shape, more than colour or texture

Template theory

Match to a stored shape; problem: you'd need too many templates

Recognition by Components (Biederman, 1985)

Recognise objects by identifying basic 3D shapes (geons) and how they are arranged; works across viewing angles

Geons

Basic 3D shapes in Biederman's structural description theory

View-based theories (Tarr)

We store specific views; rotating an unfamiliar view makes recognition harder

Logothetis

IT neurons respond best to learned views, within 40° of them

Structural vs view-based debate

Unresolved; probably both are used depending on object type, complexity and familiarity

Visual agnosia

Can't derive meaning from what is seen; low-level vision intact and other senses work (e.g. Kevin Chappell)

Prosopagnosia

Selective loss of face recognition

Acquired prosopagnosia

Caused by right occipito-temporal damage

Congenital/developmental prosopagnosia

Often undetected, with compensatory strategies; sometimes misdiagnosed as behavioural problems or autism

Face perception vs face memory

Distinct but overlapping abilities

CFPT

Cambridge Face Perception Test; not all people with developmental prosopagnosia and a memory impairment fail it

Cherry's dichotic listening task

People notice pitch, loudness and sometimes their own name in the unattended ear, but not content or language

Inattentional blindness (gorilla study)

58% missed the unexpected event

Change blindness (flicker paradigm)

A blank between two images means we have to search deliberately

Broadbent's filter model

The unattended channel is filtered out early and completely

Treisman's attenuator model

The filter only reduces the unattended signal; explains the cocktail party effect

Top-down attention

Deliberate, goal-driven, effortful, slow; serial, limited capacity; e.g. serial visual search, flicker paradigm

Bottom-up attention

Automatic, reflexive, fast; parallel, high capacity; e.g. pop-out, cocktail party effect

Contingent capture

A salient or task-relevant distractor pulls attention

Pinto et al.

Performance on top-down and bottom-up tasks doesn't correlate, so they appear to be independent systems

Where does bottom-up selection start?

Parietal cortex (LIP)

Where does top-down selection start?

Frontal cortex

Value-driven attention

Stimuli once paired with reward still capture attention more than six months later

Parietal lesions

Cause visual field neglect

ADHD circuits

Fronto-striatal and fronto-parietal circuits

Attention spotlight

Attention to locations works like a spotlight, with a "zoom lens" update

Overt attention

Attention with a gaze shift

Covert attention

Attention without a gaze shift

Posner cueing

Valid cues speed responses (80% of trials); invalid cues slow them (20%)

Cue-target timing: 0ms

No benefit

Cue-target timing: 100-150ms

Peak facilitation

Cue-target timing: 300-500ms

Inhibition of return (IOR)

Inhibition of return (IOR)

Slowed response to a previously cued location; thought to aid foraging, but naturalistic searches show people don't revisit locations less

Feature integration theory (Treisman)

Brain detects basic features in parallel, then uses focused attention to bind them into objects

Binding problem

How separately detected features are combined into a single object

Single-feature vs conjunction search

Single-feature searches pop out; conjunction searches get slower as distractors increase

Exceptions to conjunction search slowing

Three-feature conjunctions can be easier; motion conjunctions can pop out

Multitasking

Really rapid task switching; switch costs are reduced by preparation and predictability but not eliminated

Types of driving distraction

Visual, motor, cognitive

Cognitive distraction while driving

Causes visual tunnelling; worse for novices because hazards appear in the periphery

Signs and ads while driving

Capture attention at the expense of driving information

Attention training after TBI

Training on alertness, vigilance, selective and divided attention helped, especially the intensity functions (alertness and vigilance)

Action video gamers

Greater spatial attention capacity, larger useful field of view, better temporal resolution, less fronto-parietal activation for the same task

Real-world benefits of action gaming

Linked to benefits in surgery, UAV piloting and simulated driving