4 Representation of Complex Images Flashcards
Where is the Dorsal Stream and the Ventral Stream in the brain?
Dorsal stream, from visual cortex along the top of the brain. (contributes by specifying movement used for grasping a target)
Ventral Stream, from visual cortex along the bottom of the brain. (provides information about object size, shape and texture).
What happens to neurons as the angle of the face seen changes?
Different neurons will respond to different angles of the face.
What is a Jennifer aniston cell? and what does it tell us?
A cell that just response to Jennifer antison.
Tells us that it is not just a female face cell that appears, but a cell that responses to a particular person, not anything else.
What does the Halle Berry Cell tell us?
Suggests that the concept of halle berry is enough to trigger the cell. Therefor it is a high conceptual response.
What does “invariance” suggest about cells?
That cells can encode a representation so that it is identified regardless of the size, orientation, colour etc
What does the cell found about the Sydney opera house suggest?
That invariance is not just for faces, but architecturally discriminating as well. Familiarity is important.
Could cell firing be modulated by familiarity?
A neuron in the amygdala had a selective excitatory response to an image of just one of the research running an experiment.
Not an emotional response as it does not response to spiders etc
What is a grandmother cell (gnostic)
A cell that only fires when you see your grandmother. So when these cells are fire it is your neruons firing that tell you this is a person. only requires a small amount of cells.
Problems with the grandmother cell (local coding) mechanism
cause Everything you can identify as seperate from something will need its own cell group., therefore is is susceptible to damage. If it got damaged you would not be able to recognise your grandma. This is not true because brain decreases gracefully when damaged.
- Generlization is difficult
- Pattern completion requires access to representations of other similar objects (need all parts of grandma to know its her)
Critical goals of visual information processing
- Seperate patterns
- Complete Patterns
- Allowing for generlisations (so we don’t get fucked up by tigres)
Pareidolia?
Completing images
What is the trade off between between seperation, pattern completion and generlisation?
The nervous system does this. And balance will likely produce some intermediate coding scheme where representations are distributed but maintain some degree of sparseness.
Recent Evidence for a distributed Facial representation code
Face stimuli were deconstructed into points and shading. And then synthetic faces were made from averages.
They found that cells are turned to one axis of face space and are blind to changes on the axis.
They found you can decode from cell activity. Like going into the mind of the person.(predict what face were presented solely from knowing what image fired).
