Theories Of Vision Flashcards
what are the three theories of vision
marr - computational theory
gibson - ecological theory
Bayesian Inference
define Marr’s computational theory
visual system works by info processing
3 tiered system
1. theory - define input and output, and establish constraints
2. algorithm - use a specified logical/mathematical algorithm to yield an output from input
3. implementation - nerve cells carry out algorithm
give an example for 1. computational theory
binocular stereopsis (3d vision) input = image from eyes, output = depth due to view as midpoint from both eyes contraint - images correspond to only one scene point
give an example for 2. algorithm
stereoalgorithms - light/dark boundaries, course/fine detail as input to compute depth
ie. very blurred likely to be CLOSE
give an example for 3. implementation
‘disparity sensitive neurons’ - used in stereopsis to implement depth
what is marrs heirachy of representations
image - primal sketch - 2 1/2 D sketch - 3D model - action
primal sketch and 2 1/2 D sketch seperates imageinto its seperate components ie geometry, reflectance, illumination and viewpoint
define raw primal sketch
first in hierachy of representations
detects changes in intensity, geometric structures and illuminations
ie. gausian blurring
what does the raw primal sketch map
important primatives within an image
ie blobs - small areas bounded by changes in intensity
edge segments - sudden changes in intensity
bars - two parallel edge segments
termination - sudden discontinuity
define blobs (RPS)
small areas bounded by changes in intensity
define edge segments (RPS)
sudden changes in intensity
define bars (RPS)
two parallel edge segments
define termination (RPS)
sudden discontinuity
define full primal sketch
after raw primal sketch, primatives are grouped and sumarised to construct a larger and more abstract geometic illusion
produces a hierachy of primatives ie hairs > stripes > texture
presents location, shape, texture and details
define 2 1/2 D sketch
occurs during full primal
imput of info about slants and distance - look at depth relation, distance between obecs and observer to work out how far an object extends
what does 2 1/2 d sketch use to determine depth
motion cues
texture cues
disparity between two eyes