Exam 2 Flashcards
Spatial Coding Systems
allocentric: object-to-object
egocentic: self-to-object
Pohl
functional organization of visual pathway
monkeys – lesioned posterior parietal cortex (PPC) and inferior temporal cortex (IT)
landmark discrimination –> spatial discrimination (food hidden in adjacent well to landmark; correct=retrieve food, incorrect=empty well shown to convey no reward)
object discrimination –> choose food well
PPC: important in landmark task, not important in object (spatial, visual relationships, dorsal)
IT: important in object task, not important in landmark (object and pattern recognition, ventral)
Dorsal
occipital–>parietal
“where” pathway; spatial relations
Ventral
occipital–>temporal
“what” pathway; object recognition
Object recogniton
process of matching representations of organized sensory input to stored representations in memory
The binding problem
If ventral-stream neurons encode stimulus identity but not location, and if dorsal-stream neurons encode stimulus location, but not identity, how is this information brought together to create the unitary, “bound” percept that we experience of an object occupying a location in space?
Binding criteria
Nearness of lines
Nearness of color
Coherent motion
Experience!
Rules of object recognition
How do we go from edge detectors (in V1) to knowing what edges go together?
proximity connectedness color closure continuation
Colinearity v. relatability
Colinearity - orientations are similar
Relatability - easy to connect one line to the next
Gross experiment (hand)
waved hand in front of display screen –> high response
neuron’s preferred stimulus: hand
other stimuli could drive response based on hand similarity
responded regardless of location (large receptive field)
visually receptive IT neurons responded to complex stimuli
- preferential to complex stimuli
- large receptive fields
an IT neuron with responses selective to a hand
Type and Token
type: category (ex: faces)
token: example within this category (ex: specific face of someone)
Gross face cells selective for type, not token
Gross Experiment (face)
Single neuron recorded in temporal cortex
Neuron likes faces (category or type) - but not any particular face (token)
responds comparably to human and monkey faces
gradually less responsive as features are removed
did not respond to stimuli that wasn’t face
Gnostic cell
hypothetical neuron that represents a complex but specific concept or object
Issues with grandmother cell theory
Problems: 1.Need a lot of neurons!
2.Response ambiguity
vulnerability of system that relies on highly specialized neurons on the apex of the processing system–damage?
how would the system a priori know how many gnostic cells are required to represent every distinct object it would acquire throughout its life
Hierarchy of stimulus representation
bridging gap between V1 and IT/STP (superior temporal gyrus)
complete object recognition
component shapes
conjunction of features
low level features
progressively higher levels of stimulus representation are constructed at progressively higher levels of the system by selective integration of more elemental information from lower levels
Lines and colors represented by V1 neurons are assembled by higher visual areas into recognizable objects
Filling in contours
Visual illusions provide evidence for top-down influences
Zurich
find selective neuron
present illusory contour
lower, but still selective response
inference from v4 fed back to v2
aperture problem
each neuron with a small receptive field is, in effect, viewing the visual scene through a very small aperture
view invariance
recognizing objects irrespective to viewpoint
particular shape invariant - location - size - cue (color, motion, lines, texture)
Visual agnosia
not knowing through visual information
–> cannot experience perception of an object
limited to vision
ex: tactile modality intact (holding keys)
object knowledge is okay
Visual prosopagnosia
inability to recognize faces
Template matching theory
image generated by a stimulus is matched to internal representation (template)
works well when object is well specified and unique
Challenges for template matching
imperfect matches
not powerful enough for general pattern recognition
ex: many fonts of m
cannot account for flexibility of pattern recognition system
Feature matching theory
detect objects by the presence of features
each object broken down into features
ex: broken down A
These 3 features are in most As Line features activated by visual cortex
Stored representations
Stored representations ≈ features that are relatively common to all instances of object, and relatively rare in non-instances
Problem to feature matching
many objects contain similar features
doesn’t work for faces – same features
Recognition by components
Biedelman
geon model
Complex objects are made up of arrangements of basic, component parts; 24 of them
Tanaka experiment
how specific do neurons get?
critical features
- -> isolate single neuron
- -> present monkey with dozens of 3D objects to find driving cell
- -> reduction process until neurons no longer fire
not responding to specific objects, instead tuned to a simpler, reduced set of generic features
tiger’s head is recognized by the simultaneous activation of many neurons that each represent
What’s special about faces
Recognition of con-specifics critical for survival
Faces seem to be recognized by configuration
Inversion has more detrimental effects on faces than on other classes of objects
Faces recognized as individuals
Thatcher effect
more difficult to detect local feature changes in an upside-down face, despite the same changes being obvious in an upright face
when face is rotated away from upright, adults see it as decreasingly bizarre
tuned especially to upright faces
differentiation depends heavily on configuration (the structural relationship between individual features on the face)
Configural processing
Determine extent to which quantitative spatial relations deviate from prototype (average)
Recognition based on “distance” between perceived item and prototype
Faces differ in relative sizes of parts and distances between them
Evidence for configural processing
Famous faces: better at recognizing caricatures than veridical drawings
caricatures make deviations from prototype more evident
Inferotemporal cortex
IT
Kanwisher experiment
FFA
area in fusiform gyrus much more active response to face stimuli
Face inversion
face-specific processing system that can be accessed only by upright faces
parts of the face are not processed independently
prosopagnosia
Configural processing
perceiving relations among the features of a stimulus such as a face
contrasted with ‘featural processing’
Composite face effect
Evidence for configural processing
Subjects are slower and less accurate in recognizing the top half of one face presented in a composite with the bottom half of another face when the composite is upright and fused
can detect the difference when the composite is inverted or the two halves are offset laterally
–> disrupts holistic processing
This phenomenon demonstrates that when upright faces are processed, the internal features are so strongly integrated that it becomes difficult to parse the face into isolated features
Caricatures
Evidence for configural processing
Caricatures
Participants’ recognition of facial expressions was enhanced when differences between locations of features in an expression face and a reference-norm face (e.g., neutral face) were accentuated
The exaggeration of these deviations in a caricature may enhance recognition because it emphasizes the features of the face that are encoded.
Face superiority effect
2 noses vs. 2 faces differing only in nose
Better discriminations when whole face present! (and better memory for the noses presented with faces)
face superiority effect disappears when inversion occurs
parts of the face are not processed independently
recognizing nose less accurate than recognizing larry’s face
Evidence for configural processing
Recognize faces based on spatial relations between features
Composite face effect (can’t tell difference when fused, can tell difference when holistic processing is broken)
Caricatures (easier to recognize bc encoded features are exaggerated)
Inversion (face-specific processing system that can be accessed only by upright faces)
Thatcher (more difficult to detect local feature changes in an upside-down face)