Object + Face Recognition

Question 1

Features detectors:

Answer 1

To process objects we use patterns recognition

Simple feature detector neurons in visual cortex respond to features such as lines, dots, colours etc

Question 2

Patterns recognition

Answer 2

We sometimes process whole scene before perceive details

Our brain = wired to put individual features together automatically

Study found: preformance speed with small letter slowed if large letters were different

Question 3

Perceptual organisation : what’s the law of proximity?

Answer 3

visual elements close in space tend to be grouped together

Question 4

Perceptual organisation: what is law of similarity ?

Answer 4

Similar elements are grouped together

Question 5

Perceptual organisation: what is law of continuation?

Answer 5

We group together elements requiring fewest changes in straight/ curving line

Question 6

Perceptual organisation: what is law of closure?

Answer 6

Missing parts of figure= filled in to complete it

Question 7

Figure/ ground:

Answer 7

Remainder = ground
Figure= main part of visual fields

Face-goblet illusion

Question 8

Gestalt psychology: what is law of Prgnanz?

Answer 8

We perceive simplest possible organisation of visual field

Question 9

What do gestaltsist assume about figure-ground segregation?

Answer 9

Does not depend on past experiences / learning

Question 10

Evidence against gestalts theory of figure- ground segregation

Answer 10

Ppl suffering with amnesia presented with various stimuli

Some containing parts of well-known objects

Amnesic patients shows no diference in 2 types of stimuli
Healthy ppl could identify regions with familiar objects

Question 11

What does FIRGURE-ground segregation rely on?

Answer 11

Does not solely depend on basic features
Also depends on past exeprive > object familiarity (amnesiacs struggled ,normal ppl dont)

Question 12

Spatial frequency theory:

Answer 12

We have 2 systems:

1. We quickly process low spatial frequency (LSF)
2. Then process high spatial frequency milliseconds later (HSF)

Question 13

Spatial frequency + ventral/ dorsal streams

Answer 13

LSF info processed by fast M (magnocellular) Pathway (rods, V1, dorsal)

HSF processed by P (parvocellular) Pathway (cones, ventral, V1)

Question 14

Spatial frequency: Mona Lisa

Answer 14

Spatial frequency explains why Mona Lisa has elusive smile
Smile more obvious in LSF images

Our central vision = dominated by HSF

Different neurons respond to HSF than LSF
- explains why we experience scenes as a whole, then can focus in on details

Question 15

Object recognition theories: Marr’s computational approach

Answer 15

• argued object recognition involves various processing strategies

Primal sketch:
- provides 2-dimensional description of main light intensity changes in visual input > includes info about edges

2 1/2 D Sketch:
- description of depth + orientation of visible surfaces
- makes use of info from shading, texture, motion etc
- observed centred

3D model:
- describes 3-dimensional objects shapes + their position independent of observes viewpoint

Question 16

What did Marr realise?

Answer 16

Object recognition= more complex that originally thought

Developed comprehensive computational models of processes involved in object recognition

Distinctions between viewpoint- dependent and viewpoint-invariant

BUT TOO FOCUSED ON BOTTOM-UP PROCESSING

Question 17

Biederman’s recognition by component theory

Answer 17

• Geons= basic shapes + components
• states viewing from any angle would still lead to similar object recognition (viewpoint invariant)
• ## importance of edges in object recognition

Question 18

Biedermans 5 properties of edges:

Answer 18

Curvature > points on curve
Parallel > sets of points in parallel
Lotermination > edges terminate at common point
Symmetry
Collinearity > points sharing common line

Question 19

Biederman’s steps in object recognition:

Answer 19

1. Edge extraction- various aspects of visual stimulus processed
2. Then object segmented into geons

Question 20

Support for Biederman’s theory:

Answer 20

Single neurons in IT is macaques respond to geons regardless of angle

Others say he emphasised importance of edges

Question 21

Top- down processing

Answer 21

Barr (2006)
- presented particpants with drawings of objects that were masked
- activation in orbifrontal cortex occured 50 ms before activation in recognition
- top-down process in orbifrontal cortex = more important when recognition is difficult

Question 22

Top down process (triggered by verbal labels) activated shape info + influences basic visual detection

Answer 22

• observed told stimuli would be circle or square
  -then presented with circle, square or no shape
  -prefomance = better when valid cue given

Question 23

Top down process and allocation of attention

Answer 23

Top-down process influences allocation of attention
Then allocation of attention influences bottom-up processing
Knowledge drives search for features
Repeated until recognised

Question 24

Object recognition conclusion

Answer 24

How we organise perception helps us understand basics of object recognition
LSF proceeded parallel to HSF
Marr and Biederman theories emphasised bottom-up processing for unambiguous objects
Top-down becomes more important with degraded images

Question 25

Face recognition : how do computers recognise faces?

Answer 25

- maps geometry of face - distance between eyes - distance from one forehead node to eyebrows - 68 measures - compare facial signatures to all signatures on data base

Question 26

Face recogntion: how d humans recognise faces ?

Answer 26

We don’t know exactly Holistically: We look at whole face to recognise not specific features Holistically processing of faces =more efficient + reliable than feature processing More rapid and reliable than object recognition Facial features processed in parallel, rather than 1 by 1 Familiar faces can be recognised in under 1/2 second

Question 27

Face inversion effect:

Answer 27

We’re slower at identifying faces when inverted Evidnece for holistic processing Evidnece suggests this doesn’t apply to object recognition Dog experiment showed inversion effect for dog faces Car experiments had much smaller inversion effects

Question 28

Face blindness: prosopagnosia (pross)

Answer 28

- not all ppl show facial inversion effect - ppl with pross dont have impaired facial processing - some have intact object recognition - shows facial and object recognition = different systems Pross= heterozygous Can be caused by brain damage Shows evidnece for covert recognition - face processing without awareness

Question 29

Prospoagnosia study eg:

Answer 29

-Ppl with proso presented with familiar + unfamiliar faces - proso showed more activation in brain area associated with face processing when presented with familiar faces - familiar faces processed below level of conscious awareness

Question 30

Why do propagnosiacs have poor facial Recognition?

Answer 30

Have face-specific recognition disorder - damage to brain area specialised for face processing Bad at identifying complex stimuli

Question 31

Propagnosia: object and face recognition

Answer 31

Can have normal levels of object recogntion, but poor facial recogntion Objet recog: - increasing degree of similarity between targets increased error rates Sam in normal ppl and propagnosiacs Facial recognising: - propagnosiacs preformed poorly, even wheat easy for normal people

Question 32

Double dissociation

Answer 32

If face and object recogntion involve different brain areas, should have patients with impaired object + intact face recogntion >>> rarer Study : man with agnosia (poor object recogntion) preformed same as healthy ppl on face recogntion

Question 33

What is the Fusiform Face area?

Answer 33

Area associated with face processing Partners with propagnosia have damage to this area Study: Only 80% of ppl showed greater activation in fusiiform area to faces than objects

Question 34

Differences between face and object recognition?

Answer 34

Face: - more holistic processing > face inversion effect - faster due to parallel processing Object: - less holisitic > smaller inversion effect - slower: mix of parallel + serial processing