15. object recognition Flashcards
1
Q
2 approaches for object recognition?
A
- recognition by components
- viewpoint-specific
2
Q
Recognition by components?
A
- use geons to assemble and recognize a whole object
- A model of object recognition that proposes that recognizing an object depends on first identifying the object’s basic 3-D shapes and how they fit together.
3
Q
Evidence for recog. by componenets?
A
- we can rec. an object even if we only have some of its geons
- its really difficult to identify an object if you can’t identify geons
4
Q
Limitations of recog. by components?
A
- it achieves invariance… but its often too abstract
- ALL briefcases would look the same, couldn’t DISTINGUISH between them
- the model is too extreme
5
Q
viewpoint-specific meaning?
A
- objects are recognized based on view-specific representations stored in memory
- ie. multiple representations, for each possible view, are stored at the same time
6
Q
Evidence for viewpoint-specific?
A
- artificial objects are recognized more quickly when seen from a familiar viewpoint
7
Q
Limitations for viewpoint-specific?
A
- do we have enough room in our brains to store all the different representations we would need?? (no)
8
Q
Grandmother cell?
A
- truly invariant –> responds to an image of a toaster and the concept (word) toaster
- can’t be used to guide action
- debated, cuz it would require enormous amounts of neurons for all objects and concepts
- A neuron that responds to a particular object at a conceptual level, firing in response to the object itself, a photo of it, its printed name, and so on.
9
Q
Top-down meaning? What kind of information?
A
cognitive influence on perception
- perceiver’s goals
- attention
- prior knowledge
- expectations ab what objects are likely to occur in the current scene
10
Q
Bottom-up meaning?
A
based only on stimulus
11
Q
How does the “gist” affect object recognition? Vice versa?
A
- We’re much worse at identifying objects when they’re at an atypical location in the scene or in an inappropriate scene
- We’re much better at identifying the correct scene (background) if the object and scene are correct
12
Q
How are recognition of the gist and the object related? —> When?? What helps what?
A
- we likely process both at the same time
- after identifying “gist” –> top down expectations of what objects are most likely to be in certain scenes help with object recognition
13
Q
Bayesian approach?
A
- visual system uses 2 probabilities to infer the scene from the retinal image:
1. prior probability of all scenes
2. for each scene, probability that it produced the current retinal image - use of mathematical probabilities to describe the process of perceptual inference
14
Q
Spatial frequency def? What does high vs low look like?
A
- describes periodic distribution of light and dark in an image
- low: fuzzy, big blobs of color
- high: only fine details, little/no color
15
Q
Units for spatial frequency?
A
- Cycles per degree
- more cycles = higher sp. frq.
