What is perception?
a cognitive process that uses our previous knowledge to construct, identify, and interpret the stimuli that our senses register
Why we need perception?
�for object recognition
�for action
the inverse problem
detecting a signal away from its source, then trying to determine or reconstruct the source
(light detected by receptor cells at the retina forms a 2D image, while the real world is in 3D! - present in all perceptual modalities)
distal stimulus
object in the world that emits or reflects energy
proximal stimulus
�pattern of energy that reaches sensory receptors
perceptual object
what we perceive as a result of our sensory activity and post-sensory processing
from retina to brain
retina optic nerve lateral geniculatenucleus (LGN) primary visual cortex (V1) higher visual cortices association cortices
the likelihood principle
given all the possible stimuli that could produce the proximal stimulus, assume the most likely one to be the distal stimulus
frame of reference
apply the likelihood principle
in context!
in perceiving brightness and color what are 5 assumptions we make?
- surfaces are uniformly coloured
- gradual changes in brightness are caused by shadows
�light comes from above
- any light source producing a shadow is stationary
�local contrast
monocular depth clues: pictoral cues: 6
1. occlusion
2. linear perspective
3.atmospheric perspective
4. texture gradient
5. height-in-plane relative size
6. familiar size
convergence:
binocular depth dues: the extent to which thetwo eyes are turned inward to fixate an object
stereopsis
Binocular Depth Dues: the visual perception of depth due to:
�retinal disparity-the difference betweenwhat the two eyes see(where on the two retinasan image projects).
retinal disparity
the difference betweenwhat the two eyes see(where on the two retinasan image projects). (the closer the object,the larger the retinal disparity)
What is perception?
a cognitive process that uses our previous knowledgeto construct, identify, and interpret the stimuli that our senses register
top-down processing
perceptual and associative knowledge acquired from past experienceguides our current perceptions.
bottom-up processing
assumptions and cues we use to go from sensory input to percept are learned!
ventral pathway goes from..
from visual cortex (V1)
to inferior temporal cortex (IT)
dorsal pathway goes from..
from visual cortex (V1) to posteriorparietal cortex (PPC)
ventral stream is considered the _______pathway
what pathway : object recognition
(conscious) perception of the visual properties and identity of objects; view-invariant
dorsal stream is considered the _______ pathway
where pathway (also how): location and movement for action
(unconscious/automatic) visual guidance of movement; egocentric
lesion to the _____ causes visual form agnosia
inferior temporal lesions : ventral stream
lesion to the _____ causes optic ataxia
posterior parietal lesions ; dorsal stream
visual form agnosia
unable to recognize object size/shape/orientation. BUT able to use visual features to guide precise hand and finger movements (finger-thumb grasp)
the dorsal stream must have access to information about the visual features of the object (e.g. shape and or
THE VENTRAL STREAM extra DIDN'T GET IT...
single neuron recordings in monkeys:
�connectivity:
V1 V2 & V4 inferior temporal (IT) cortex
(feedback connections too!)
�each area represents the entire visual field
�neurons in these areas progressively increase
the sizeof their receptive fields
the com
v1 is the...
primary visual cortex
selectivity of v1 neurons: findings were...
cells in the primary visual cortex are selective for line orientation, direction,and speed of movement
what are the two models of visual recognition (ie larger theories) their definitionand their subcatagory models
1. viewer centered theories:
the perceptual representation of the object depends on the viewpointof the perceiver
�template-matching model
�feature-matching model
2. object centered theories
the perceptual representation of the object is independent of th
what are the two viewer centered theories?
template-matching model
feature-matching model
template-matching
must find an exact object-level match
problem: very rarely will you see the exact same object from the exact same perspective twice!
changes in viewpoint produce drastic changes in the proximal stimulus
feature-matching
match some collection of individual features
problem: deals with size variance but how about orientation variance? being egocentricis a problem!
�how do you know which are the critical features?
changes in viewpoint produce drastic changes in the proximal
biedermans geon model
recognition by components
what are geons
geometric ions
�36 simple shapes
�distinguishable from almost every viewing perspective
�recognizable even when parts are obscured
�fundamental for object recognition
�building blocks of every object
geon model: objects are defined by... (2)
objects are defined by:
�their constituent set of geons
�the spatial relationship between those geons
geon model: object recognition (2)
1.extract appropriate geonfrom image
2.match to similar representations stored in long-term memory
16
geon model experiment
vertices -where two or more line segments come together
-are critical for the detection of geons
prediction:
�if the vertices are deleted, object perception is impaired; if they are not, it's normal (even if slower)
does viewpoint matter?
usually can recognize objects at even the oddest of viewing angles, visual identification is fastest from certain viewpoints.
very small differences in object properties can make a world of difference!
most objects have a canonical view - what is this?
the most common viewing angle; the view at which recognition is fastest and most accurate
results of canonical object test:
familiarized subjects with novel objects at one viewpoint, and tested recognition at different viewing angles
recognition was slower the more the object is rotated from its learned viewpoint
viewer based model
whole-object views are stored in memory
not ALL possible viewpoints, only the most common (canonical) ones
in matching-to-template, the current object representation can be mentally rotated/zoomed
face recognition: findings
found that participants were better at identify whole faces than parts
also found that this part/whole difference did NOT hold for other kinds of stimuli: scrambled faces, inverted faces, houses
concluded that faces are represented holistically-parts are
what is the FAA
the fusiform"face" area - in right inferior temporal cortex
when damaged leads to prosopagnosia-a profound selective deficit in face recognition
two views: no there is no FAA because
faces are only different from other objects because we identify them not at the superordinateor subordinate level but at the specificlevel.
faces are represented holistically because this is how expertsrepresent their objects of expertise.