Exam 2 Deck Flashcards
Exam Feb 27
Agnosia is
Inability to recognize visual objects with intact visual acuity
Agnosia results from
Results from damage to inferior temporal cortex
Contralateral Hemifield Neglect
Inability to attend to contralesional space
Contralateral Hemifield Neglect results from
posterior parietal lobe damage (usually right
side)
Effect of Lesions in the Monkey: Inferior Temporal Cortex
Food hidden under object
• Monkey must select proper object to get reward
– Must have intact object recognition system • Can not do with inferior temporal lesion
• Can do with parietal lesion
Effect of Lesions in the Monkey: Posterior Parietal Cortex
• Food hidden in well farthest from peg
• Monkey must choose location farthest from peg to get reward
– Must have intact spatial relationships system • Can not do with posterior parietal lesion
• Can do with inferior temporal lesion
Two Visual Pathways Beyond V1
- -> Dorsal “Where” spatial location stream. located in posterior parietal
- -> Ventral “What” object feature stream located in inferior temporal
Law of Proximity
Things that are close together belong together
Law of Similarity
Things that are alike get grouped together
Law of Good Continuation
Things that result in straight or smoothly curving lines, rather than abrupt angles, get grouped together
– We see a curved line over a straight, not two lines with abrupt angles
Law of Common Fate
Things that move together get grouped together
Closure
We group things that close
Common Region
Things that are enclosed together belong together
– Can overcome proximity
Connectedness
• Things that are connected to each other belong together
– Can also overcome proximity
Illusory Contours
- Gestalt laws of perceptual organization and figure/ground principles are illustrated by seeing illusory figures
- By altering features of picture elements, illusory occluding objects can be perceived
Grouping Principles and the Kanizsa riangle
• Law of Good Continuation
– Straight lines and smooth angles are more likely than sharp
corners
• Law of Simplicity
– The simplest explanation is 3 disks and 2 triangles rather than 3 ‘hats’ and 3 ‘pac-men’
– So that’s what the visual system perceives
What’s the Figure & What’s the Ground?
• We can find the edges
– Separates the green from the blue • But is one thing in front of another?
– Probably see green objects on a blue background
Occlusion Determination Heuristics
• A “heuristic” is a ‘rule-of-thumb’
– A rough rule that works most of the time • Two occlusion heuristics
– Relatability – Non-accidental features
Relatability
• Edges that can be connected with a simple curve (e.g. elbow) are “relatable”
– Likely belong to the same object, even if occluded
– Like Gestalt good continuation
• If we see relatable features, we’re more likely to perceive that
something is occluded
Non-Accidental Features
• When 3D objects overlap, they create some specific kinds of junctions that don’t vary by viewpoint
– Y or ‘arrow’ junctions • Corner
• Not an occlusion – T junctions
• Intersecting edges • Occlusion
T/F: The perceptual system tries to find an interpretation that depends on an accident of viewpoint
false, it tries to find an interpretation that would be consistent across most viewpoints, so one that does not depend on accident
Resolving Perceptual Ambiguity
• Early vision
– Pretty simple
• Dots and bars
• Not much question about whether a dot is there or not
• Middle vision – More complex
• Which edge belongs to what object • Not always clear
How Does the Perceptual System Decide Which is the Best
Interpretation?
• A Perceptual System Metaphor – Decision by committee
• Multiple members – Some cooperate – Some compete
• Make decisions
– e.g. this corner goes with this edge – These edges belong together – This is the figure, this is the ground
Pandemonium: A Perceptual Committee Model
• Committee members (“demons”) at multiple levels, shouting
– Feature demons
• “shout” louder when they think their feature is present
– Cognitive demons
• “shout” louder when they think their object is present
– Decision demon
• Listens to the noise and determines who is shouting
loudest
• That’s the representation that ‘wins’
T/F: perceptual “committees” honor physics
true, they don’t come up with any decisions that violate physical reality and don’t settle on an interpretation that has rocks floating in
the air
T/F: perceptual “committees” do not avoid accidents
FALSE, they discount interpretations that rely on rare or unlikely conjunctions of features
How Do We Recognize Objects?
• Template matching?
– Have a template for all objects – Extract a set of features
• Edges
• Corners
– Find out what’s in front of what
• Figure & ground
– Try fitting the figure into a bunch of templates until you find
the one that fits
• When the figure fits, like a piece in a puzzle, we’ve
identified the object
What are some problems with templates?
• Object variance
– The same object can have a lot of different exemplars • Dining char
• Rocking chair
• Recliner
– Each exemplar can be seen from many perspectives • Front
• Side
• Back
• We’d need way to many templates for each object for this to
ever work
What are Geons?
– Components from Biederman’s structural theory
• “Recognition by Component”
– Every object description consists of geons in specific, non-
accidental configurations
What is Prosopagnosia?
• A face-specific agnosia
– Inability to recognize faces
• Implies specific neural area for a special kind of object
recognition
– Located in inferior temporal cortex – Part of the ventral ‘what’ pathway
• Some researchers think there is no such thing as prosopagnosia, but that we have a special ‘expertise’ area, and that we’re all facial experts
Basic Principles of Color Perception
• Color
– Not a physical property of matter
– Rather a psychophysical property of the perceptual system • However, based on properties of matter
– Most of the light we see is reflected
• Typical light sources: Sun, light bulb; emit a broad
spectrum of wavelengths 400–700 nm
– Different kind of materials absorb and reflect different
wavelengths of light
Are there “basic” colors?
Evidence for basic colors – Color sorting studies – Color words across cultures
what are the four primary colors?
– Red
– Blue
– Green
– Yellow
What did Newton discover with the prism?
• When white light passes through a prism it is decomposed in to component parts (different wavelengths) of different colors
What wavelength of light is perceived as violet?
400-450 nm
What wavelength of light is perceived as blue?
450-500 nm
What wavelength of light is perceived as green?
500-570 nm
What wavelength of light is perceived as yellow?
570-590 nm
What wavelength of light is perceived as orange?
590- 620 nm
What wavelength of light is perceived as red?
620-700 nm
What is the first hypothesis of photoreceptor response?
Hypothesis 1:
– A single photoreceptor that responds differently to different wavelengths could index an object’s color
What are some problems with the photoreceptor hypothesis?
Univariance. Different wavelength-intensity combinations can elicit exactly the same response from a single type of photoreceptor
– One type of photoreceptor cannot make color discriminations based on wavelength
• We must have multiple color receptors
How many colors can we see?
somewhere between 2,000,000 & 10,000,000 different colors we can perceive
We have 10,000,000 different types of color receptors, one for each of the 10,000,000 possible colors.
FALSE, we have 3
Hermann von Helmholtz discovered …
Discovered that we have three color spectral sensitivity curves
– Blue
– Green
– Red
Young-Helmholtz Trichromatic Theory
• We have three color receptors
• These receptors are sensitive to different parts of the EM
spectrum
• Differential relative activity in these receptors is the basis of
color vision
Trichromatic Theory
• Color perception is based on differential activity in the three receptors
– Depending on the energy of light in different frequency bands
• Comparing output of the three receptors allows color computation in the brain
What are S-cones?
Cones that are preferentially sensitive to short wavelengths (AKA ‘blue’ cones)
What are M-cones?
Cones that are preferentially sensitive to middle wavelengths (AKA ‘green’ cones)
What are L-cones?
Cones that are preferentially sensitive to long wavelengths (AKA ‘red’ cones)
T/F: Each wavelength in the visible spectrum will result in a unique response pattern across the receptors
TRUE
T/F: Light usually comes to our eyes as pure wavelengths
FALSE, all light is a mixture of wavelengths
T/F: Reflected light has some wavelengths absorbed but still a mixture of wavelengths reflected
TRUE
What two colors are mixed together to perceive yellow light?
red and green
What are metamers?
• Different combinations of wavelengths of light that produce the same perceived color
– 550 nm + 650 nm gives same color percept as 590 nm light • Thus they are metamers
– Note that these are not ‘blends’
• Not like mixing chocolate & coffee to make mocha
– The individual wavelengths are still there
• They just activate the receptors in the same way • Results in the same color percept
Combing lights is ________
additive color mixing
Combining pigments is ________
subtractive color mixing
Additive Color Mixing
• Combine short wavelength light
– Appears blue
• With combined middle and long wavelength light
– Appears yellow
• Get short, middle, and long wavelength light
– Appears white
Subtractive Color Mixing
• Instead of blue and yellow light, take blue and yellow paint
– Top paint patch absorbs (i.e. subtracts) most long wavelengths reflecting mostly short and some middle
• Appears blue
– Bottom patch absorbs mostly short wavelengths reflecting
mostly middle and some long • Appears yellow
Color Subtraction
• Mix the blue and yellow together
– Subtracts the long and short wavelengths
• What’s left unsubtracted is the middle – Appears green
• This is how painting mixing works
What can trichromatic theory not explain?
- Contrast effects
* Afterimages
What are contrast effects?
– Same wavelengths
– Same activation pattern
– Different color percepts
What are afterimages?
-No color in the stimulus
– Color in the percept
• Afterimages demonstrate opposition color pairings
– Stare at red, get green afterimage (and visa versa) – Stare at blue, get yellow after image (and visa versa)
What was Ewald Hering’s “opponent process” theory?
• Knew about after images
• Also noted that color-blindness often was paired
– e.g. red/green color blind, rarely yellow/blue but never yellow/green or red/blue
• Noted that some color combinations were never described – e.g. you can have ‘bluish-green’ or ‘yellowish-red’ colors, but
not ‘reddish-green’ or ‘bluish-yellow”
• Proposed three opponent color mechanisms
– Green-off, Red-on – Blue-off, Yellow-on – Black-off, white-on
Receptors are _______
trichromatic.
- Three types of cones sensitive to different parts of EM spectrum
Retinal ganglion and LGN cells are _________
Opponent Process.
– Color center-surround organization
Explain what is used for the Mondrian experiment
• The Mondrian – A large screen with patches of different colors • Three pure wavelength light sources – Long (red light) – Medium (green light) – Short (blue light) • Telephotometer – Measures the intensity of reflected light – Aim at a patch on the Mondrian
Explain the Mondrian experiment
• Aim photometer at a patch on Mondrian (e.g. a green one)
– Turn on long wave light
• Adjust intensity until photometer reads “60” units reflected
from the patch; turn off – Turn on medium wave light
• Adjust intensity until photometer reads “30 ” unites reflected; turn off
– Turn on short wave light
• Adjust intensity until photometer reads “10 ” units
reflected; turn off
• Turn all three lights on: Ask subject “what color is the patch?”
– Spectral content of light reaching subject’s eyes = 60 long, 30 medium, 10 short
– Subject reports patch color is green
Explain the repeat of the Mondrian experiment
• Aim the photometer at a blue patch
– Adjust light sources so the SAME wavelength reflectance is measured as before, but now from the blue patch
• Spectral content at eyes reflected from blue patch = 60 long, 30 medium, 10 short
– Ask what color the subject sees
• Subject reports “blue” (even thought the wavelengths
reflected from the patch reaching the eye are EXACTLY
the same as from the green patch)
• Do the same for a red patch, subject reports “red”
What did the Mondrian experiment find?
• In all cases the spectral content of light at the retina reflected from the patch was the same
– Long: 60; Medium: 30; Short: 10
• Yet the perceived color was different
– Red, green, or blue
• Thus, while perceived color is related to light wavelength, it is
not determined by wavelength – Something else contributes
Explain the “void” experiment
• Set light reflected from green patch equal (e.g. L:30; M:30; S: 30
• Restrict the field of view so that only that one patch is visible – Subject reports color is grey
• Allow to see all patches, keeping light constant – Subject reports patch color, e.g. “green”
• Spectral content from patch is same
– Percept ‘pops’ back and forth from gray to green
What did the “void” experiment find?
If only one reflectance surface is in the field of vision, color perception is entirely determined by wavelengths of light reflected from the surface
• If more than one surface if available, perceived color is dependent on both the reflected wavelengths from the surface and the reflected wavelengths from the surrounding surfaces
T/F: The percentage of a given wavelength of light reflected by something is always the same
TRUE
Define: Biological Color Separation
• Any scene will have a different light/dark brightness pattern for each of the three cone types
– The light/dark pattern is different for each receptor • it’s a color separation
– These light-dark patterns are the same, independent of the illuminant, only the relative intensity changes
• Thus the color system can ‘correct for the illuminant’
What is Land’s Retinex Theory?
• The visual system compares the reflectance (brightness) records across all the surfaces (colored patches) for each cone
– How much ‘blue’, ‘green’, and ‘red’ brightness there is
– Gives an estimate of the spectral content of the illuminant
• The visual system then compares the light/dark patterns at a
given patch for each cone
• The results of the two comparisons are combined to obtain a
color perception, corrected for illumination
• This is a computation (and a kind of complicated one), not a
simple reflection of the world
Where does the color process happen in the brain?
Color is part of the photopic system
- cones –> Parvo but then a slightly different pathway
V1 Color Areas
• In V1, color information is processed in “pegs” or “blobs” in the hypercolumns
– The initial brightness records are probably assembled here
Higher cortical color processing
• The final comparison of brightness records probably occurs in cortical area V4 in prestriate cortex
– In the monkey V4 neurons are the first ones to respond to perceived color rather than wavelength of light
– Lesions to human V4 leads to achromotopsia, the inability to see color
Define: Euclidian geometry
– Parallel lines remain parallel as they are extended in space – Objects maintain the same size and shape as they move
around in space
– Internal angles of a triangle always add to 180 degrees – etc.
T/F: Brain must reconstruct Euclidean geometry from two flat, distorted projections from the retinas to create a 3D image.
TRUE, the brain has to answer the question “what is in front of what?” from a flat picture.
What are some depth cues in a 2D image?
• Monocular cues – Pictorial cues
• Static cues in 2D representations – Movement cues
• Dynamic cues in 2D representations • Oculomotor cues
– Physiological feedback from the eyes
• Binocular cues – Binocular disparity cues
• Derived from the differences in images on the two retinas
Monocular (Pictorial) Depth Cues
• Occlusion (things in front block things behind)
• Relative size (how big compared to scene)
• Texture gradient (equally spaced elements get packed in the
distance)
• Relative height (proximity to horizon)
• Familiar size (how big compared to knowledge)
• Atmospheric (aerial) perspective (haze)
• Linear perspective (convergence)
Define: Occlusion
A cue to relative depth order when one object obstructs the view of part of another object
How do we know that there is depth?
• The Retinal Image is Ambiguous
– First do object recognition
– Then figure out what’s in front of what
• Could be top object arrangement – Violates Gestalt principles
• Bottom more likely
– Circle in front of square in front of triangle – Conforms to Gestalt principles
What do nonmetrical depth cues do?
Provides only qualitative information about the depth order
(which thing is in front of what) but not depth magnitude
(how far in front the thing is)
What do metrical depth cues do?
– Provides quantitative (how far) information about distances between objects
T/F: Image size can provide a depth cue regardless if we know how big it actually is.
FALSE.
The farther away something is, the smaller its retinal image is, thus image size can provide a depth cue but only if we know how big something is
Relative Size
• A comparison of size between items
– Smaller items appear more distant
– If the things look the same, we can assume they’re about
the same size
– Therefore, smaller ones are more distant
– Don’t have to know the absolute size of any of them
T/F: If the different sizes are randomly scattered, rather than along a gradient, relative size is a stronger cue.
FALSE.
If the different sizes are organized along a gradient, rather than randomly scattered, relative size is a stronger cue
What is Atmospheric Light Scattering?
Haze
• Air scatters light
– Results in haze
• The perceptual system ‘knows’ this
– Uses haze as depth cue
• Distant items dimmer and fuzzier than close items
– Also bluer since short wavelength light scatters more • That’s also why the sky is blue
Atmospheric Perspective
- Relative height without a texture gradient gives only a weak (or no) depth perspective
- Relative height plus haze gives a stronger depth perspective
Linear Perspective
• Parallel lines appear to converge in the distance
– Ultimately converging at the vanishing point
• The visual system uses this as a depth cue
Define: Motion Parallax
– Near objects appear to move farther than far objects as we go by them
– The retinal projection of a near object travels farther across the retina than a far object for the same movement
Define: Deletion and accretion
– Changes in occlusion due to movement
Define: Optic flow
– The distance on object is from us alters its apparent
direction and speed of movement
Oculomotor cues
• Feedback from eye muscles
– Convergence
• The movement of the eyes towards “crossed” to foveate
near objects – Accommodation
• The change in lens thickness (fattening, mediated by ciliary muscles) to focus on near objects
• These cues are only effective out to ~10 feet
Binocular Depth Cues
• Binocular disparity cues
– Derived from the differences in images on the two retinas
T/F: Animals with eyes on side have overlapping visual fields
FALSE. Only animals with forward facing eyes have overlapping visual fields
The Vieth-Müller circle
Images on the retinas of items on that circle are the same distance from the fovea on both eyes
– They fall on corresponding retinal points
The Horopter
• Extend that circle so it’s a dome in front of the eyes equidistant as fixation and you have the horopter
– A 3D surface of corresponding retinal points
The Horopter and Corresponding Retinal Points
• The circle (or 3D dome) with its center halfway between the eyes and the fixation point is the horopter
• Items on the horopter project to corresponding retinal points on the two eyes
– Equal distances between the objects retinal projection and the fovea
Binocular Disparity
• Objects on the horopter (including the fixation) have zero binocular disparity
– Images fall on corresponding retinal points
• Objects off the horopter have binocular disparity
– Images fall on non-corresponding retinal points – Unequal distances between the fovea and the object’s
projection on the retina
• The farther off the horopter the object is, the larger the
disparity
• Binocular disparity is the binocular depth cue
– Lets the perceptual system know, from two flat projections, how close or far away from fixation distance the object is
• The greater the disparity, the farther from fixation
Crossed & Uncrossed Disparity
• Retinal image projections with the same amount of disparity can have either crossed or uncrossed disparity
– Crossed disparity
• Retinal projections are outside of fovea • You’d have to ‘cross’ your eyes to focus on it • Closer to you than fixation
– Uncrossed disparity
• Retinal projections are inside of fovea
• You’d have to ‘uncross’ your eyes to focus on it • Farther from you than fixation
What is Stereopsis?
• The perceptual phenomenon of depth – Things appear to ‘pop-out’
• As opposed to flat appearing
– e.g. 3D movies or static images with stereoscopes/colored
glasses/polarizing glasses • Or the real world
Angle of Disparity
• Angle between P & F and P’ & F’
– Disparity angle = 0 for points on horopter
– The farther off the horoper an object is, the greater the angle
of disparity
Disparity-Sensitive Neurons
- There are neurons sensitive to different degrees of binocular disparity
- These neurons give rise to stereopsis
Binocular V1 Neurons
• Some visual cortex neurons have binocular input (instead of being monocular dominant)
– These binocular neurons respond best at specific angles of disparity (this one likes 30’)
Disparity Tuning in V1
• Most neurons respond best to zero (or near zero) disparity
– In the fusion area
• However, some respond best to disparity between retinal
images
Bayesian Theory
- P = Probability
- Sx = Scene X
- I = Perceptual system input
Limited Capacity & Selection
• The brain is a limited capacity information processing system
– It cannot process all of the available perceptual information – Must somehow select a subset of available percepts for
additional processing
– This selection is called “attention”
Intuition vs. Definition
• We may know what attention is intuitively, but it’s hard to define objectively
• Not one thing but a collection of selection processes – selection by novelty
– selection by relevance
– selection by location
– selection by feature
T/F: Participants are fastest on valid trials because of facilitation
true, they are slowest on invalid trials because of inhibition
T/F: How much time there is between the cue and the target changes the reaction time effect
TRUE
Objects Play a Role in Attention
• You get an advantage by paying attention to the object in which the target appears
• You pay a cost by paying attention to the other object
• Even though the distance between the cues and the targets is
the same – Same distance
Cues Aren’t Required for Attention
• Selection by Relevance
– We attend to things relevant to our current tasks – Even if they’re not changing
Treisman’s Feature Integration Model of Visual Search
- The visual system has primitive feature maps – Maps of the locations of primitive features in space
- These are perceptual maps created and maintained in parallel – i.e. you create a color map and a separate orientation map at the same time and they are independent of each other
Search in the Treisman Model
• Preattentive stage
– If target is based on a single feature, we can simply poll the
correct map and ask “is the feature present”
• e.g. is there anything in the ‘vertical’ orientation map or
the’red’ color map
• Don’t need to know where it is • It just “pops out”
• Efficient search
• Attentive stage
– If the target is defined by a conjunction of features, the
attention ‘spotlight’ must scan the objects in the scene to see if the target defining features exist in the same place on the relevant feature maps
• e.g. are ‘red’ and ‘vertical’ in the same location • If so, then the target is present
• Inefficient search
What are these Primitive Features?
- Orientation
- Length/Width • Size
- Curvature
- Closure
- Density
- Color
- Intensity
- Intersection
- Flicker
- Motion direction
The Binding Problem
- Even if features are in the same location, they must be ‘bound together’ to create the proper percept for target identification
- We still don’t understand binding, but we can confuse the system
The Speed of Attention
• Attending to spatial locations has a temporal course – Rising for ~150 ms
– Falling back to zero until ~300 ms
– Inhibition until ~700 ms
• What about object attention?
Testing Attention Speed
• Rapid Serial Visual Presentation (RSVP)
– Present stream of items, one at a time, at fixation, very
quickly (e.g. one every 100 ms)
– Make items distinct from one another (e.g. letters and
numbers)
– Have participants respond to one category of the items (e.g.
the numbers) while ignoring the others • Modified RSVP
– Have two targets instead of one
• e.g. respond to the digits 3 and 7 and ignore all the other
letters and numbers
– Vary the amount of time between successive targets
The Attentional Blink
- If the targets occur between about 200 and 300 ms of each other, the participant will miss the second target after getting the first
- It’s as if their attention selection system ‘blinked’ for a moment following correct identification of the first target
The ‘Fishing’ Metaphor of the Attentional Blink
• Percepts are like things flowing by in a (dirty) river – Branches
– Old boots
– Tires
– Fish
• We’re fishermen
• Attention is our net
• We can monitor all the things going by until we ‘catch a fish’ – Respond to a target
• Responding to a target takes some time
– During that time our ‘net is out of the water’, and we’ll miss
any other fish (targets) that come along