Lecture 12 - Object recognition, part 2 Flashcards
dorsal pathway
how/where
ventral pathway
more complex, identity pathway
infero-temporal cortex
identification
fusiform gyrus
fMRI studies in humans found that it was special involved with FACIAL processing
You perform a series of single-cell recordings and find that a small set of neurons consistently fire in response to pictures of
Ryan Gosling. What form of representation best describes this
result?
sparse coding because it’s a SET of neurons
Domain specificity models
let’s look for places in the brain that deal with the important stuff on an evolutionary standpoint
Suggests that certain brain regions are dedicated to specific categories (domains) of objects [e.g. faces, animals, tools,
etc.]
- Nativist view – more modularity, less plasticity
- Expect to find sparse codes, keep the numbers confined to a specific area
property-based models
almagom of different types of knowledge, lots of context associated with the object
- Claims that object knowledge is linked to sensory and motor attributes of the item [note that this is more than the physical features].
- Constructivist/embodied cognition view – less modularity, more plasticity (whatever a group of neurons is responding to can change what it’s going to respond to over time if trained correctly)
- Expect to find population codes.
fMRI experiments that localize face processing
Greeble
Experience dependent
plasticity in humans
fMRI experiments show that training tunes FFA (fusiform face area) neurons to respond to novel or non-face stimuli.
• In the Greeble recognition study,
FFA was found to respond to artifical creatures following 7 hours of training.
- Neurons in FFA have also been shown to respond to cars for people who have expertise in identifying cars.
- This is taken as evidence for the property-based theories.
prosopagnosia
domain people like this
damage to FFA, from stroke, tumor, or other injury
this is the inability to recognize/distinguish faces - even your own
however, it is only a visual deficit. hearing a voice can allow identification
additionally, prosopagnosia can occur when you have damage to other areas (so no double-dissociation)
other areas also involved with facial recognition
superior temporal sulcus, inferior occipital lobe, temporal pole, and frontal areas
sometimes we see faces that aren’t there
pareidolia
generalized phenomenon of perceiving a vague stimulus as something of importance (seeing jesus on a grilled cheese)
the visual system may be biased to create (recognize) a perception that is ambiguous in the stimulus
this is the general case, we construct our own reality all the time. we don’t have direct access to that distal stimulus and the neural architecture is putting it together
The parahippocampal gyrus
(known as the parahippocampal
place area or PPA)
responds best to spatial layout (including buildings and places).
some specialization to knowing where things are
Extrastriate body area (EBA)
how we represent bodies
specialized to bodies, doesn’t really care about the face
it will respond strongly to stick figures, parts of the body (arm flexing)
won’t respond at all if things are reconfigured, it’s not the color, the contrast, the lines, it’s really the body
in occipitotemporal cortex, responds best to pictures of full
bodies and body parts (not faces).
Binocular Rivalry: Experiment by Tong et al. (1998)
demonstrated?
MRI using binocular rivalry, two overlaped images on a screen that are diff colors, person where colored eye glasses, so the two different images go to different eyes
same distal and proximal stimuli stays the same: but you’re only aware of one of those images at one time: flips back and forth
– Binocular rivalry was used - each picture shown to one eye at the same time.
– Picture of a house shown to one eye and a face to another. You can only perceive (i.e. are aware of) one at a time.
– Participants pushed button to indicate perception and recognition
– fMRI showed an increase in activity in
• Parahippocampal place area for the house
• Fusiform face area for the face
– Demonstrates how perception and recognition correlate with specific brain activity. = = = > brain parsing out this perception task
Can’t we build a model that brings these levels of explanation together? That is, can’t we determine how features (and their biological implementations) combine to represent all objects?
Nope.
because how do we get from moving bars to faces
object perception is hard
- The stimulus on the receptors is necessarily, logically ambiguous.– Inverse projection problem
- Objects can be hidden or blurred.
– Occlusions are common in the environment
- Objects look different from alternative viewpoints
– Viewpoint invariance:
– This is (still) a difficult task for computers to perform
Inverse projection problem:
any object, any proximal stimulus that you can have on your retina can be produced by an infinite number of things in the environment
distal things can be different but they can produce the same pattern that the receptors respond to
An image on the retina can be caused by an infinite number of objects. [Many different distal stimuli can create the same proximal stimulus.]
Viewpoint invariance
it doesn’t matter really what angle you’re looking at an object you can still identify the object
the ability to recognize an object regardless of the viewpoint/perspective.
Notice that all these ‘problems’ assume you are trying to find…..
…..whole objects, not features or properties
pattern matching, that you have a picture in your head and your trying to match
basic properties and relationship of features might be more important
bias the system
top down
which of these things are what i expected to see
Structuralism was established by Wilhelm Wundt (late 1800s)
what sort of approach to perception?
what couldn’t structuralism explain?
a sensation as a basic perceptual unit: very hierarchical
– States that perceptions are created by combining elements called “sensations.”
– Hierarchical, well-ordered approach to perception. Strongly “bottom-up.”
– Structuralism could not explain apparent movement.
apparent movement/motion
when motion is perceived but not actually there
• Motion is perceived when separated objects (e.g. dots, bars, etc.) flash in different locations, but without the necessary sensory atoms predicted by structuralism.
