Topic 6: Objects and Scenes Flashcards
inverse projection problem
2d image on retina from 3d world
different objects can look identical
single object can look different
viewpoint invariance
ability to recognize an object regardless of viewpoint
-object constancy
gestalt approach
principle of good continuation
whole is other than the sum of parts
- principles of perceptual grouping
- lines tend to be seen as following smoothest path
principle of good figure/simplicity/pragnanz
principle of similarity
stimulus patters are seen so the resulting structure is as simple as possible
-similar things appear grouped together
principle of proximity/nearness
principle of common fate
nearby objects appear grouped together
-elements that move together appear grouped together
principle of common region
principle of uniform connectedness
principle of perceptual separation
- elements that are in the same region of space appear grouped together
- a connected region with the same visual elements is perceived as a single unit
- allows us to separate elements apart
figure-ground separation
figure is more thing-like and memorable than ground
-figure is in front of ground
ground is more uniform and extends behind figure
border ownership-contour separating figure from ground belongs to figure
-convex shapes are usually considered figure
holistic (gestalt)
characterized by comprehension of the parts of something as intimately interconnected and explicable only by reference to the whole
recognition by components theory (RBC)
we perceive objects by perceiving elementary features
geons = 3d volumes
an object is recognized when enough info is available to ID geons
discriminability
resistance to visual noise
invariance
GEONS
-geons can be distinguished from other geons from almost all viewpoints
-geons can be perceived in noisy conditions
-recognizable no matter the illumination direction, surface markings or texture
36
principle of componential recovery
-key to object recognition is not amount of info but ability to ID components or geons
-quantified by amount of surface area and amount of lines covered
critical juncture point determine ID ability
JIMS model
layer 1
layer 2
neurally inspired, multi-layered, bottom up
starts with visual scene and travels up for processing
-basic visual features (edges and orientation)
-gives info of higher processing and more detail of environment (long lines and vertices)
gist of a scene
according to RBC theory we build up from small details to overall object
-experiments on scene perception suggest we can perceive large properties first (250ms) and then fill detials
degree of naturalness
degree of openness
undulating corners vs straight lines
visible horizon vs close environment
degree of roughness
degree of expansion
colour
large even areas vs many small elements
-convergence of lines vs parallel lines
characteristic colours
experience and perception
top down
perception based not just on the stimulus but depends on experience context goals
letters or numbers perceived based on orientation
figure-ground separation experiment
when black image is similar to shadow it is more likely to be seen as figure
-when ambiguous black and white equally likely to be seen as figure
meaning and familiarity influences separation
likelihood principle
unconscious interference
we perceive the object most likely to have cause the pattern of stimuli we receive
-application of likelihood principle is unconscious but based on past experience
visual (what) stream
Fusiform face area
parahippocampal place area
extrastriate body area
-ventral stream and object ID
-responds to images of faces
-responds to images of places, houses scenes (bottom of temporal lobe)
responds to images of body parts and bodies in motion, in between dorsal and ventral stream
visual expertise in FFA
task is to recognize faces or made up greebles
before training ffa was higher for faces
training was matching following blocking
after training there was no difference in ffa activation between faces and greebles
ffa can recognize similar objects that are distinct
PPA vs 3d space
space defining objects and space ambiguous objects
SD objects activated PPA more - responds more than FFA but still not fully defined
reconstructing visual experiences from brain
teach and use decoder
reconstruction
record brain activity while P watches trailers, build regression models, build library to put through regression and generate predicitons
-record brain activity for new set of trailers, select 100 clips which is similar to brain activity
average of clips togther
neural decoding of neural imagery during sleep
train linear support vector machine (SVM) on fMRI data measured while each person viewed web images
-record fmri during sleep, predict images and compare to verbal descriptions
higher accuracy with higher levels of visual processing stream
ex. ffa more accurate than v1
