w4 receptive fields & convolution

Question 1

what is a receptive feild (RF)?

Answer 1

the area on the retina where light can influence the firing rate of a neuron (e.g. retinal ganglion cell)

Question 2

how are receptive fields organised in the retina?

Answer 2

in concentic zones either ON-centre, OFF-surround or OFF-centre, ON-surround, responding in opposite ways to light

Question 3

what happens when light hits the centre of an ON-centre cell??

Answer 3

firing rate increases (excitation)

Question 4

what happens when light hits the surround of an ON-centre cell??

Answer 4

firin rate decreases (inhibition)

Question 5

what happens in an OFF-centre cell when the centre is illuminated?

Answer 5

firing rate decreases (inhibition)

Question 6

what is meant by spontaneous firing in RCGs?

Answer 6

baseline neural activity (~50 impulses/sec) even without stimulation.

Question 7

what ahppens when the surround of an OFF-centre cell is illuminated??

Answer 7

firin rate increases (excitation)

Question 8

why is spontaneous firing important?

Answer 8

it allows both increases and decreases in firing rate to signal different kinds of information (excitation/inhibition)

Question 9

why does the visual system use ON and OFF cells equally??

Answer 9

to efficietnly detect both increases and decreases in luminace -> enhancing contrast detection

Question 10

what do retinal ganglion cells respond to best??

Answer 10

contrast (differences in luminance), not uniform brightness

Question 11

what happens when uniform light covers the entire receptive field??

Answer 11

excitatory and inhibitory effects cancel -> no significant change in firing rate.

Question 12

why do RGCs act as edge detectors??

Answer 12

they fire most when the centre and surround are differently illuminated -> at edges or borders

Question 13

what is the effect of RF size on visual processing?

Answer 13

smaller in the fovea (fine detail, high acuity); larger in the periphery (coarse detail, high sensitivity)

Question 14

what creates the RF’s centre-surround structure?

Answer 14

converging excitatory inputs via horizontal/ amacrine cells in the surround

Question 15

Why is this RF’s centre-surround structure useful??

Answer 15

it reduces redundancy and emphasises useful structure like object boundaries.

Question 16

What is a step edge?

Answer 16

A simple transition from light to dark (e.g., object boundary).

Question 17

How do ON-centre RGCs respond to a step edge?

Answer 17

Strong firing when the bright region falls on the centre but dark region on the surround.

Question 18

How do OFF-centre cells respond to the step edge?

Answer 18

Opposite pattern to ON-centre RGCs — they fire where light intensity decreases.

Question 19

What happens when cells move across an edge?

Answer 19

A distinct pattern of increased and decreased firing rates — highlighting the edge’s location.

Question 20

Why is this edge coding important?

Answer 20

It forms the basis for detecting shapes and structures in the visual scene.

Question 21

What is convolution?

Answer 21

A mathematical operation combining an input image with a weighting pattern (the receptive field) to produce a filtered output (neural image).

Question 22

How is convolution performed biologically?

Answer 22

Each neuron multiplies the light intensity within its receptive field by its weighting values, sums the results, and outputs a response.

Question 23

What do the weights represent?

Answer 23

Positive (excitatory) values in the centre and negative (inhibitory) values in the surround.

Question 24

What does the output of convolution represent?

Answer 24

A neural image that encodes edges, contrasts, and luminance changes.

Question 25

What are the benefits of convolution in vision?

Answer 25

Edge detection Redundancy reduction Data compression Efficient representation of structure

Question 26

How does the retina perform convolution?

Answer 26

By having many RGCs with overlapping receptive fields, each filtering different image locations simultaneously.

Question 27

What is the Hermann grid illusion?What causes the illusion?

Answer 27

A visual phenomenon where grey spots appear at the intersections of a white grid on a black background. Greater surround inhibition at intersections where ON-centre cells receive more total light → reduced firing → perceived grey.

Question 28

Why do the grey spots disappear when you look directly at them in the Hermann Gird Illusion?

Answer 28

Foveal receptive fields are smaller → less surround inhibition → no illusion.

Question 29

What does the Hermann grid show about vision?

Answer 29

Perception is based on computational filtering in the retina, not direct photoreceptor activity.

Question 30

What happens when excitation and inhibition are perfectly balanced? What is the advantage of this balanced system?

Answer 30

No change in firing → indicates uniform illumination. The visual system can ignore redundant uniform information and focus processing power on meaningful features.

Question 31

What happens when excitation and inhibition are unbalanced?

Answer 31

The neuron signals a contrast change — encoding an edge.

Question 32

What is spatial filtering?

Answer 32

Extracting specific spatial features (edges, textures, shading) by processing different spatial scales.

Question 33

What shape describes receptive field weighting functions?

Answer 33

Gaussian (bell-shaped) profiles — strong in the centre, weaker toward the edges.

Question 34

What does a broad Gaussian filter detect?

Answer 34

Coarse, low-frequency information (large, gradual luminance changes).

Question 35

What does a narrow Gaussian filter detect?

Answer 35

Fine, high-frequency information (sharp edges, small details).

Question 36

How many spatial scales does the visual system operate across? Why are multiple spatial scales necessary?

Answer 36

Roughly six scales, with ON and OFF systems for each → about twelve neural images. Real-world images have structure at different sizes — using multiple scales captures both global shape and local detail.

Question 37

What is a zero-crossing in neural response?

Answer 37

The point where convolution output changes from positive to negative (or vice versa) — marking an edge.

Question 38

Why are zero-crossings important?

Answer 38

They define object boundaries and guide shape recognition in higher visual areas.

Question 39

What is a “neural image”?

Answer 39

The pattern of ganglion cell outputs after convolution — a map of contrast-based information.

Question 40

Does the brain “see” a neural image directly?

Answer 40

No — it’s part of the internal code that higher areas interpret to construct perception.

Question 41

What is multi-scale processing?

Answer 41

The brain analyses the image through filters of different sizes simultaneously to extract features of various spatial frequencies.

Question 42

What type of cells correspond to different scales?

Answer 42

Small receptive fields: parvocellular (P) cells — fine detail. Large receptive fields: magnocellular (M) cells — coarse structure/motion.

Question 43

How does this parallel processing benefit vision?

Answer 43

Provides both high-resolution details and global scene structure at once.

Question 44

How can damage to retinal ganglion cells affect vision?

Answer 44

Loss of contrast sensitivity, reduced edge detection, and visual field deficits.

Question 45

How does the Hermann grid illustrate retinal processing?

Answer 45

It shows how centre-surround inhibition creates perceptual artefacts, proving active processing at the retinal level.

Question 46

Light → retina → RGCs

Answer 46

transduction.

Question 47

RGCs → LGN → V1

Answer 47

transmission & coding.

Question 48

RFs detect??

Answer 48

contrast

Question 49

Convolution =

Answer 49

biological filtering operation.

Question 50

ON/OFF-centre cells =

Answer 50

edge detectors

Question 51

Hermann grid =

Answer 51

lateral inhibition illusion.

Question 52

Multiple spatial scales =

Answer 52

coarse + fine analysis.

Question 53

Gaussian filters =

Answer 53

mathematical model of RF weighting.

Question 54

Zero-crossings =

Answer 54

edges in neural image.

Question 55

Efficient coding =

Answer 55

less data, more meaning.

Question 56

Bottom-up =

Answer 56

data-driven processing.

Question 57

Top-down =

Answer 57

contextual feedback.

Question 58

Retina + LGN =

Answer 58

early computational vision.

Question 59

Multi-scale RFs →

Answer 59

basis for shape & pattern recognition in cortex.

Question 60

Perception =

Answer 60

interpretation of neural codes, not direct seeing