1
Q
why is colour important?
A
aids discrimination and detection
Important in many key tasks:
- When choosing what to eat.
- Scene segmentation.
- Visual memory.
- Mating rituals.
- Camouflage.
2
Q
what is Hue (H)?
A
the quality that distinguishes red from blue, i.e., the hues of the rainbow
3
Q
what is Brightness (V)?
A
the perceived intensity of light (sometimes lightness).
4
Q
What is saturation (S)?
A
characterizes a colour as pale or vibrant.
5
Q
what is colour?
A
- subjective
- objects appear coloured because they reflect different wavelengths of light from different parts of the visible spectrum
- a property of our neural apparatus - need to have the correct photoreceptors and neurons to see colour
6
Q
what is a metamer?
A
- sensory stimulus that is perceptually identical to another stimuli, but physically different (e.g. a light that appears orange is indistinguishable than a combination of red and yellow light)
- suggests the visual system is producing identical neural responses to physically different stimuli
6
Q
how is colour coded in the retina?
A
in the photoreceptors the cone cells and their photopigments properties
7
Q
what are the 3 cone types?
A
- S cones (short λ, blue); peak absorption at 420nm.
- M cones (medium λ, green); peak absorption at 530nm.
- L cones (long λ, red); peak absorption at 565nm.
8
Q
what is the principle of univariance?
A
- how one can discriminate between wavelengths through comparison of multiple photoreceptors
- If the intensity of λA is the same as λB, then there will be a different response from the cell to the different light
- But if the intensity of λA is about 2x the intensity of λB, then the response from the cell will be the same to both.
9
Q
how does wavelength discrimination improve?
A
- with the number of clone classes
- dichromats = 2 pigments
- humans = trichromats (3 cone types S, M, L)
10
Q
what is opponent coding theory?
A
colours are grouped into opposing pairs (blue and yellow, red and green)
10
Q
what is retinal topography? (cone mosaic)
A
layout of cone cells on the retina
1. There are far fewer S cones (blue) than M or L
2. There are no S cones in the fovea
3. They are randomly distributed, but clumping is common.
4. The layout and relative proportions of cones is largely individual, e.g., some will have roughly equal amounts of L and M comes, while others will have a L:M ratio of 4:1.
11
Q
what are Parvocellular RGCs?
A
P-type retinal ganglion cells project to the parvocellular layers of the lateral geniculate nucleus.
12
Q
what is the physiology of opponency?
A
- Parvocellular RGCs have chromatically opponent RFs (centre-surround antagonism)
- have ON and OFF versions
- e.g. The centre may be excited by red light, while the inhibitory surround is excited by green light
13
Q
how does colour tuning work in the LGN?
A
- LGN layers 1 & 2 get their input from M RGCs: input for achromatic luminance channel.
- Layers 3-6 get theirs from P RGCs: input for the two chromatic channels, called cardinals.
14
Q
how does colour tuning work in the visual cortex?
A
- cortical cells show a preference for a wide range of hues, not just the cardinals
- Tuning width remains fairly consistent across cortical areas (V1, V2, V3).
15
Q
what is colour constancy?
A
the ability to assign a fixed colour to an object even though the actual spectral information entering the eye changes in different illumination conditions
16
Q
what is acquired colour vision definicy?
A
(cerebral achromatopsia) is typically due to damage to V4.
16
Q
what is congenital colour vision definicy (CVD)?
A
- an X-linked recessive gene
- XY chromosomes: 8% chance of colour blindness. XX chromosomes: 0.5% chance of colour blindness.
- affects M or L cones, rather than S cones
- If M or L cones are missing, then green and red will be confused.
- If S cones are missing, blue becomes hard to distinguish.
17
Q
what is many-to-one mapping?
A
- when many separate objects that occupy the same cognitive category
e.g. Every single one of these is a rocking chair
Every single one of these is a letter A.
18
Q
what is the two-stream model?
A
Post V1, information is transmitted via two pathways.
- Ventral Stream: the ‘what’ pathway.
From V1 to V2, then V4 and IT cortex. Associated with object recognition, and memory.
- Dorsal Stream: the ‘where’ or ‘how’ pathway.
From V1 to V2 , then V6/DM and finally V5/MT.Associated with motion, location, saccadic control.
19
Q
what is object agnosia?
A
damage to the ventral stream which causes a deficiency in object recognition
20
Q
what are the models of recognition?
A
- template-matching models
- Feature detection models
- Structural description models
- View dependent models
21
Q
what is the template-matching models?
A
- simplest form of object recognition
- you have a detector (e.g. for the letter A)
- when an object appears in the RF of this detector that matches this template it signals
- for this to work we needs a detector for every possible orientation, scale and font
22
what is the feature detection models?
Selfridge’s Pandemonium model (1959) - described it in terms of demons with different jobs (sticking with the letter example)
1. The feature demons look at the image and simply write down how many examples of their feature they see (e.g., tuned to horizontal lines).
2. The cognitive demons shout if they think that combination of features applies to their letter; the more confident they are, the louder they shout.
3. The decision demon listens to the cognitive demons and decides who is shouting the loudest, providing that as the perceived letter.
23
what is the structural description models by Marr & Nishihara?
- goal of the model to describe object unambiguously
- system must be invariant to transformation viewpoints → system must know which properties are invariant under transformation, and how other properties might vary.
1. Object-centred negates the problem of transformation variance.
2. Volumetric approach: volumes only require axis and size info – maintains specificity without requiring too much storage space.
3. should be organised into an object description - description modular and hierarchical, can be described at many scales, allowing for identify matching and discrimination.
4. Recognition: the "model store"
- even if doesn't match exactly you'll Find the closest match and Have sufficient information on the object from the image and your memory to help you interact with it (image manipulation)
24
what is the structural description model, Bidermann?
- proposed a set of primitive volumes into which objects are decomposed (not just cylinders).
- The volumes are called geons (geometric ions).
- estimated ≤ 36 of these geons
- there are 362 = 1,296 pairs of geons, which can be attached in different ways and of different relative sizes
- there are ~75,000 possible 2-geon objects
25
what are the pros of the structural description model?
- Invariance is well explained.
- Recognition relies on description rather than matching.
- Graded representations cope with discrimination and generalization.
- Evidence that structural information matters to humans and to neurons
26
what are the cons of the structural description model?
- Extracting model parameters can be hard in real images (e.g., occlusion).
- Structural description is difficult for some objects (e.g., crumpled paper, campfires).
- Driven by theoretical desirability rather than behavioural or physiological evidence.
27
what is the view-dependent model?
- Human object recognition is not perfectly viewpoint invariant.
- The viewing sphere: practiced recognizing objects from specific viewpoints (shown as black spots), tested at novel viewpoints.
- Interpolation: between previous viewpoints. Easiest.
- Extrapolation: beyond previous viewpoints but in the same axis. Medium difficultly.
- Orthogonal axis: from a completely new viewpoint. Hardest.
28
what are the pros of view-dependent model?
- Straightforward.
- Minimises transformations that must be performed.
- Newer models are based directly on what we know of physiology.
- Abstract features are recombinable.
- Good behavioural, physiological, and simulation-based evidence.
29
what are the cons of view-dependent model?
- Humans often show quite good generalisation across viewpoints even for novel objects.
- Still more memory intensive than e.g. geon model.
30
why is face perception interesting?
- uniquely rich in information such as Identity, familiarity, age, race, gender. Gaze direction, attractiveness, mood, communication
31
what is pareidolia?
when you see faces that aren't there e.g. in clouds, on toast, taps etc.
32
where are faces processed?
- The Fusiform Face Area (FFA)
- face selective region, shown in contrast studies
Evidence from physiology (Desimone et al., 1984)
- Neural signaling in monkey FFA was highest for faces (of the same species).
- When scrambled or partially obscured, the response went down
33
what is the domain specificity hypothesis ?
- faces are special
- We are born with dedicated mechanisms for facial recognition, which operate differently to those that serve typical object recognition
34
what is the expertise hypothesis?
- Faces are not special
- Face perception simply shows us how general object recognition mechanisms work for objects we are extremely well-practiced at observing.
35
Domain Specificity Hypothesis
Evidence 1: Neonatal face discrimination
- Newborn babies prefer to look at face-like patterns more than non-face-like patterns
- But this might be a broader preference for top-heavy patterns
- But babies as young as 1-4 days old seem to be able to tell their mother’s face from that of a stranger
35
Domain Specificity Hypothesis
Evidence 2: Prosopagnosia.
- cannot recognise faces
- have different gaze patterns
- Acquired: damage to occipito-temporal regions (e.g., stroke).
- Although very rarely isolated completely to faces
- Developmental: can be hereditary
- Can be very isolated to faces.
36
Domain Specificity hypothesis
Evidence 3: the Inversion effect
- Bistable ambigram face drawings. You can see the sullen police officer, but can you see the inverted face?
(British artist Rex Whistler, 1905-44).
- It’s much easier to see the second face (the surprised conductor) when it’s the right way up.
- Pareidolia is orientation specific.
- Thatcher effect
36
Domain specificity hypothesis
Evidence 4: sensitivity to facial configuration
- the inversion effect disrupts configural information more than featural
- this is evidence of holistic processing: the inability to attend to one part of the face
- further evidence: change one part (the mouth) and the whole face looks different
37
Domain Specificity hypothesis
Evidence 5: Part-Whole effect
- Sub-parts of faces are not independently recognizable (Tanaka & Farah, 1993).
- Training phase: Participants were given a face to remember, either whole or scrambled
- Testing phase: The participants were given a distinguishing task, where one thing (nose) had been changed.
- Results: Participants trained on the whole face were better at identifying the whole face. Participants trained on scrambled faces were better at identifying individual parts
- when whole face is learnt, it is processed holistically
- when trained on scrambled face parts processed individually as face-specific mechanisms not activated
38
Expertise Hypothesis Evidence 1: the effect of (un)familiarity (Jenkins et al., 2011)
we are so much better at identifying people we have already seen. Facial recognition is heavily dependent on familiarity - you have practiced identifying these particular faces
39
Expertise hypothesis Evidence 2: the "other race" effect (Shepard et al., 1974)
people are better at remembering, more accurate at matching, and can make finer discriminations amongst faces of their own race rather than another
40
Expertise hypothesis Evidence 3: object inversion in experts (Diamond and Carey, 1986)
orientation is more critical in situations where the participant has extensive practice in making subtle object discrimination
- tested experts in dog breeds on subtle differences in pictures of dogs, upright and inverted
- both dog experts and non-experts were worse at recognising faces when they were upside down
- but only dog experts were worse at recognising dogs when they were inverted
- so the inversion effect applies to all things we are good at recognising
41
Expertise hypothesis evidence 4: the part whole effect in objects (Gauthier and tarr, 2002)
parts are often recognised better in their original context, not just faces
- trained ps to recognise
Greebles
- showed them a target Greeble and told which part of it to attend to (e.g. ears), then showed that part in isolation or in its trained configuration
- was it the same or different to the part of the target greeble?
- accuracy was much better when the parts were unchanged
- so the part-whole effect exits for objects too
42
Expertise hypothesis Evidence 5: FFA activation in car experts
- FFA may be an area responding to expertise
- FFA activation as a response to faces, animals, cars and planes
- most voxels preferred faces
- but the amount that these voxels were activated by cars (vs. animals) was correlated with how expert the person was with cars
- so the FFA may aid the perception of images in which we are experts
Cognitive Psychology 2
flashcards
Decks in class (22)
# Cards
-
Introduction to Perception31
-
Taste and Smell33
-
Touch21
-
analysing images and human contrast sensitivity20
-
The Visual Pathway68
-
Visual Perception47
-
Human memory19
-
Encoding in Real World15
-
Effective encoding19
-
Storage failures20
-
Active Forgetting20
-
Types of Retrieval11
-
Improving Retrieval14
-
Autobiographical Memory14
-
Visual Word Recognition 121
-
Visual Word Recognition 234
-
Embodiment - language and reading18
-
shallow processing - language and reading21
-
Thinking 127
-
Bounded Rationality I15
-
Bounded Rationality 215
-
attention and consciousness26
