PS2021 (Perception, Attention and Action)

How do we know vision is not simple

Machines struggle with replicating human vision

How does a camera record images

Records pixel values. Records light intensity as numbers assigned to sensor locations

How does the human eye record images

Records light intensity as neural activation of photo receptors

How is the visual system like a camera

At best it is similar to the earliest stages of visual system

What is the purpose of the visual system and perception

Perception = generate meaningful and adaptive representations of the environment

Visual system = meaning and adaptive representations explicitly encode relevant aspect of the world and help guide behaviour

What is the Craik-O’Brien-Cornsweet Effect

Seeing the colours as 2 different colours when they are the same. This is due to the edge boundaries and our brain trying to make a 2D image, 3D

What does the two-tone image effect show

Initially, it looks like meaningless black and white splotches but if given prior knowledge it influences how we see it

What is the information processing paradigm

Input (stimulus)→ Brain/mind (information processing)→Output (perception)

What other disciplines is vision science important for

Psychology

Neuroscience

Evolutionary biology and sensory ecology

Engineering and machine vision

Applied mathematics and physics

What does psychophysics try to determine

The relationship between stimulus and perception quantitatively

What is the absolute threshold

The smallest amount of stimulation that can be reliably detected

What is the difference threshold

The smallest difference between two stimuli that can be reliably detected

What is just noticeable difference (JND)

Difference between guessing (getting 50% right) and knowing (e.g. 75% right)

What do thresholds typically measure

Limits of the perceptual system

Studies perception at the level of the whole organism

What is the Weber Fraction

Ratio between JND and reference intensity

The ration is always constant

What is electrophysiology

Microelectrode located near (or inside) the neuron will pick up action potentials

Provides information about processing in individual neurons or a small number of neurons

Typically done in animals (due to ethics)

What are the pros and cons of EEG/ MEG

Good temporal resolution but bad spatial resolution

What are the pros and cons of fMRI

Good spatial resolution but bad temporal resolution

Why would we use neuroimagining techniques

They can pick up consequences of electrical signals in the brain without the need for direct access

They can provide information about processing in the brain at the level of whole brain areas

What are computational methods

Build mathematical models of information processing at different levels

Provide very precise explanations (mechanistic and functional) of biological vision

Important implications for machine vision

How does vision link to neuropsychology and neuropsychiatry

Study individuals with brain lesions or with neurological/ psychiatric conditions

Important for understanding healthy vision

Allow for the creation of treatment and interventions

Two types of photoreceptors

Rods and cones

Rods

Approx 100 million

One type of photopigment

Very sensitive to light; used in dim light (Scotopic vision)

Poor spatial resolution

High temporal resolution

Cones

Approx 6 million

S-, M-, L- with peak sensitivities at different wavelengths (colour vision)

Less sensitive to light; used in bright light (photopic vision)

High spatial resolution

Poor temporal resolution

What is the receptive field

A specific region of sensory space in which an appropriate stimulus leads to a response in a sensory neuron

Structure of a ganglion cell

positive centre with negative surrounding or vice versa (equal number of pos and neg cells)

On-centre bipolar cells

Positive centre.

Response to light in the centre = depolarisation (become more positive)

Response to light in the surround = hyperpolarisation (become more negative)

Overall = excited by light shined in their centre and inhibited by light shining in their surround

Off-centre bipolar cells

Negative centre

Response to light in the centre = hyperpolarisation (become more negative)

Response to light in the surround = depolarisation (become more positive)

Overall = inhibited by light shined in their centre and excited by light shining in their surround

Eye structure

Nasal part projects contralaterally

Lateral part projects ipsilaterally

Let visual field is represented in right LGN and vice versa

Lateral Geniculate Nucleus (LGN)

Each layer gets inputs from only one eye (cells are monocular)

Arranged in a retinotopic map

The neighbouring relationships that exist in the retina, continue

Physiology of LGN

Sub cortical

Each layer contains monocular cells that form a retinotopic map of half a visual field

Layers 1 and 2 = Large cell bodies (Magnocellular layers), receive input from M ganglion cells (rod input), high contrast sensitivity and low spatial resolution, process course features and motion

Layers 3 to 6 = Small cell bodies (parvocellular layers), receive input from p ganglion cells (cone input), low contrast sensitivity and high spatial resolution, process fine features and colours

Function of LGN

First relay station between eyes and cortex (axons link to primary visual cortex)

Magno- and parvo cellular layers respond in similar ways as their input ganglion cells

Unknown the specific function

Richly connected to many other parts of the brain

A locus at which retinal information can be modulated by brain areas

More than just a relay station

primary visual cortex (V1)

LGN projects to this. Striate cortex

Left hemisphere process right visual field and vice versa

Most neurons are binocular (recieves input from both eyes)

Organised in a retinotopic map

Map is distorted (cortical magnification)

How does binocular vision help us

Helps us see depth

What is cortical magnification and why does it happen

More neural ‘real estate’ in the cortex for the centre of the visual field compared to peripheral

Causes peripheral visual to appear larger than central vision

Clarity increases the more central the object is

What does cortical magnification show us the importance of

Eye-movements

Why do we not use the same amount of neurons for the whole visual field?

Tradeoff between having a large visual field and having (some areas of) high acuity

If we would have the same acuity as in fovea in the whole visual field, we would need eyes and brains multiple times the size of our skull

Fovea

Centre of our visual field

Which receptive field do we have in ganglion cells/ LGN neurons

Centre-surround receptive field

What type of light do centre-surround receptive fields respond to

Spots of light

What receptive field do we have in V1

Elongated receptive fields

What type of light do elongated receptive fields respond to

Lines, bars and edges

If more negative cells are activated in comparisson to positive cells what happens

Inhibits. Deactivation. Do not see the light

If more positive cells are activated in comparisson to negative cells what happens

Excitatory. Activation. Sees light

What does receptor activation depend on

Orientation and size

Feature detectors

Cell responds with same firing rate to an optimally oriented but faint stimulus as to a suboptimal intense stimulus (response of single cell is ambiguous)

Bright light but not optimal orientation = faint light with optimal orientation (firing rate level)

Population codes

Takes all cells into account to make an accurate representation of light

Ocular dominance bands

Certain brain cells prefer information from the left eye over the right and vice versa.

Occurs in the V1

Stellate cells and pyramidal cells create bands of neurons that prefer information specifically from one eye

These bands are alternating (e.g. left, right, left)

Stellate cells

Receives direct input from the thalamus - specfically LGN

First cortical neurons to get visual information

Respond more to one eye than the other (ocular dominance)

Pyramidal neurons

Receive signals from stellate cells and pass them on to other brain areas

Show ocular dominance but it’s more about combining and interpreting the information

Hypercolumns

Each layer responds to a different orientation and size. Takes information from both eyes. Contains ocular dominance columns

Visual angle

Tells us the size of the retinal image (given a certain object size at a certain distance)

Fourier analysis

A way of breaking down complex patterns into its simple parts. E.g. into simple sine waves.

Early vision as local Fourier analysis

Hypercolumns contain neurons tuned to different orientations and spatial frequencies

All of these neurons analyse the same patch of visual space

Together, they extract spatial frequencies and orientations contained in 'their' local patch

Adaption in vision

method = stare at the same stimulus for a long time

Process = as a consequence of long exposure, those neurons that are tuned to the stimulus property decrease their sensitivity

Low level vision

Extracts local information about lines, bars and edges

Mid-level vision

joins isolated features into larger groups

Forms basis for high-level vision

High level vision

Object recognition

Ventral stream

‘What’ pathway = object identification

Starts in V1 and goes to temporal lobe

Receptive field size increases the further along the stream you go

Dorsal stream

‘Where’ pathway = visuo-spatial information processing

Starts in V1 and goes into parietal lobe

How does mid-level vision overcome ambiguity

Assumptions and prior knowledge

Principle of proximity

View dots as columns rather than rows due to their closeness

Principle of similarity

View it as rows due to them being separated by colour

Principle of common fate

Things that move together are bound together

Principle of good continuation

Assumes smooth (good) continuation = same object

Brain areas responsible for face perception

Inferior occipital gyri

Superior temporal sulcus

Lateral fusiform gyrus

Domain specificity account for facial processing

Mechanisms operate independently of general object perception

Expertise account for facial processing

Mechanism derive from general object perception but become finely tuned due to extensive experience

Part-whole effect

Features are easier to identify when presentedd as part of a face (e.g. eyes)

Face inversion effect (thatcher illusion)

Inversion disrupts the processing of fine details and relationship between features

As faces are upside down it is not the same process as normal facial processing

Prosopagnosia

Failure to identify or distinguish between faces, despite (otherwise) normal visual and cognitive ability

Special case of visual agnosia

Familiarity but no identification (opposite of delusional misidentification)

Norm-based code in facial processing

Facial features are represented as deviations from the average face

Makes explicit what is distinctive about a face

Emphasizes subtle variations that define individuals

if norm changes then all faces perceptions should change

Aftereffect in high-level vision

Neurons that code for facial features adapt to specific characteristics of adaptor face

As a consequence, perception of subsequent face is biased away from adaptor characteristics

Gamma rays

High energy and short wace

Radio rays

Low energy andn long waves

Visual sprectrum for humans

Between ultraviolet and infrared (approx. 400 - 800)

White light

Combination of all wavelengths

Why can’t you see colour at night

Although wavelengths stay the same, our brains ability to see differences in wavelength changes. Due to cones needing more light to activate

Rods

Very sensitive., Can react to a single photon of light

Cones

Not very sensitive and requires thousands of photons to work

What colours are we most sensitive to in lower light

Blue and green as it is when rods peak

Short cones

Blue cones

Medium cones

Green cones

Long cones

Red cones (technically most sensitive to yellow)

Principle of univariance

One cone is not sufficient to determine colour as it has nothing to compare it to. Therefore the brain compiles and compares information from all the cones to determine colour

Two receptor system

Each stimulus activates both receptors but in different ratios, and this ratio does not change with intensity. So colour can be calculated by taking the ratio of activity in the two channels

Primordial colour system

Most mammals are dichromatic (a few S cones and lots of M cones). A blue/yellow system differs to humans as our M cones are green. A purely chromatic channel with little spatial resolution

The second subsystem (colour vision)

Approx. 10 million years ago the L cones split into two. Red/green system. Only old world primates. Co-evolved with certain fruits so we could forage them

Monochromats

Only have 1 cone or just rods. Very rare. Don’t see colour at all

Dichromats

Have 2 cones

Protanopes

Deutranopes

Tritranopes

Protanopes

Lack long wavelength cone (red)

1% of males

0.02% females

Deutranopes

Lack medium wavelength cone (green)

1.2% males

0.1% females

Tritanopes

Lack short wavelength cone (blue)

Very rare

Why is colour blindness more common in males

Gene responsible is on the X chromosome. As men only have one X chromosome they only need one chromosome to have the fault whereas women need a fault on both chromosomes

Anomalous trichromacies

Cone does not peak in the ‘right’ place

Protanomaly

Deutranomaly

Protanomaly

Abnormal long wavelength cone

1% males

0.02% females

Deutranomaly

Abnormal medium wavelength cone

4.9% males

0.04% females

Cerebral achromatopsia

Colour blindness that occurs due to brain damage in the area dedicated to colour vision

What is required to see moving dots on a kinematrogram

Small displacement

Time intervals

Present pattern to same eye

Why is it surprising that we can determine depth

No dedicated part of the brain but is more intrinsic

Nothing in a photon that describes distance

Retina is basically flat so receives a flat image of the world