pattern perception Flashcards
are visual pathways heirarchial?
yes
how are image estimates transformed into known objects?
there is a sequence of processing required
what happens in the heirarchial processing sequence?
at each local patch in the image we extract info then spatial frequency tuned channels carry the signal through the early stages
what visual pathway is responsible for pattern and object perception?
the ventral cortical visual pathway
what are the two tests to object recognition?
- detection of local features (find edges)
- grouping of elements to make bigger clusters
how do we find edges?
V1 receptive fields localise points so they can find the edges
what is an issue with finding edges?
it gives us too many lines and they arent always connected around the border
how is finding edges improved?
looking for edges that coincide in different spatial frequency ranges but still need to group appropriate segments together
what does grouping involve?
contours tend to be smooth in local regions so elements that extend a contour with minimal change in orientation are likely to be the same colour
what are the gestalt grouping principles?
No grouping Proximity Similarity of colour Similarity of size Similarity of orientation Common motion Symmetry Parallelism
what do contours mark?
object boundaries which allows recognition
what needs to happen with objects overlap?
need to segment contours into objects so we used matched concavities in the profile as the point where objects are segmented
does primary knowledge have an impact on how we interpret a scene and how?
yes, contours are ambiguous so we use our prior knowledge to determine the most likely cause of the image
what is the bayesian analysis formalised expectation?
the probability of a stimulus given the image that we are seeing ie. the probability that what we think the image is is what it actually is
what are some ideas around the bayesian expectation formula?
- can potentially explain some illusions
- adaption can change expectations
- expectations often hard to accurately specify