Please enable JavaScript.
Coggle requires JavaScript to display documents.
Object Recognition - Coggle Diagram
Object Recognition
Sensory Processing
retinal image consisting of series of contours & colours
Prosopagnosia - faces
Object Agnosia - objects
Tasks in perceiving objects
Determining Boundaries
Shadows - intrinsic to the object or external lighting conditions
Image Clutter - determining the 'true' shape of an object in the world
Variable Views - Potential ambiguity of an object seen from onr viewpoint
Process - happens early on
Represent edges - Detection
Start - RGC of varying size of receptive fields
Finer Details - Smaller field; Courser details - larger field
Mexican Hat Filter
edges can be from the object or the shadow or contrast
Figure-ground
Border Ownership
Difficulty assigning boarders
Need whole image
Neural Basis
V2 neurons are pooling info from multiple V1 neurons; also determines what side the object is on as well
Many V2 neurons working, different ones for different object layouts
Grouping
Proximity
those that are closer together are grouped
Similarity of orientation
stimuli position in the same way
Common Motion
grouping those moving in the same direction together
Similarity of Colour
Similarity of Size
grouping like together
Symmetry
reflect image
Parallelism
the stimuli are parallel with each other; same shape up & down
Neural Basis
Not that these neurons are firing more or less but are two firing in the same pattern
Can be reacting to the same or different stimuli & still have a common pattern
Interpolation of missing edges
"Filling in" missing info
Illusory Contour
Edge Completion - even though there is no luminous edge present
when an edge is placed to complete the circle, we no longer fill in info
Neural Basis
firing happening when an illusory contour is very similar to that of a completed bar with edged
V2 very important
with a completed edge we get some random firing
Object Recognition - higher level processing
Match perceptual representation to those stored in memory
Neural Processing: V4 & Beyond
V1
Individual Receptive Fields
Neurons indicate presense of short straight edges at preferred orientations
V4
One large imcohmpassing receptive field
neurons indicates presence of smooth curved edge at preferred orientation
Receptive Fields
Straight or curved
Much larger than V1
Inferotemporal Cortex
Respond to combinations of contours
Respective Fields
covers the entire visual field
Responds to a certain class of objects no matter where they are in the visual field
do not have location or colour information
Use code to reduce number of neuron required
Face Area
Theories
Two Approaches centered on the nature of the representation
Difference - the number of stored images required
Prototype Matching (Structural Description Theories)
Viewpoint invariant representation - only one viewpoint required
More based on experience, collecting prototype images
Template Matching
Viewpoint dependent representation - many representations from different view points needed
accuracy comes from identifying an object in the orientation that was studied - drops when rotated
Would need a neuron for each orientation
In the our youth it may be all matching until we develop our prototypes