Chapter 4: Recognizing Visual Objects

Question 1

Tasks involved in perceiving objects

Answer 1

-Determining the boundaries of an object
>Separating out which changes in luminance and colour are intrinsic to the object or to external lighting conditions (shadows)
-Potential ambiguity of an object seen from one viewpoint (variable views)
-Determining the “true” shape of an object when some of its contours are hidden (image clutter)
-Need to be able to recognize an incredibly large number of objects in the world (object variety)

Question 2

Perceptual Organization

Answer 2

(lower level processes)
Represent Edges
Figure-ground (assign border ownership)
Grouping
Interpolation of missing edges

Question 3

Object Recognition

Answer 3

(higher level process)

Match perceptual representation to those stored in memory

Question 4

Edge detection

Answer 4

mexican mat: ganglion cell, small hat is all the increased spatial freq info and big hat is course small spac freq info

Question 5

Border ownership:

Answer 5

with the ink edges are appropriately assigned, region in front ‘owns’ the border. If slipped don’t know what’s figure-ground. Amount of light in oral field identical but firing rate for light on dark is higher. Activation is V2 higher if presented with one circle with overlapping triangles then dif vs one circle then same one. (strong response to change in border ownership)

Question 6

Grouping/ Synchronicity

Answer 6

By size proximity colour movement orientation symmetry, Single bar of light across a receptive field has high synch between neurons, two separate bars in same direction has weak sync and dif moving bars has no sync

Question 7

Interpolation with bars

Answer 7

We precieve the long light bar passed over the object so high firing since we “fill in the edge”, but if blocked off low firing since we “see” nothing pass over bar

Question 8

Prototype matching

Answer 8

(structural description theories)
Viewpoint invariant representation
Biederman (recognition by components), Marr
Based on primitives, called 3D “geons”
These are fundamental 3-dimensional shapes that are easily discriminable under a wide variety of conditions
Defines parts and the spatial relation between them
In-category distinctions difficult.
-We’d be a lot worse at object recognition if this were the case

Question 9

Template matching

Answer 9

Viewpoint-dependent representation (if change orientation we suddenly suck) therefore okay but need lots of representations for each object (unless population code, then not more neurons…more patterns of neurons)

Question 10

V4 characteristics

Answer 10

• gets input from all cells
• big receptive field
• increase complexity of what cells can respond to
• contours (roll that shape is playing in overall object)
• straight or curved
• prefered location at certain angular position

Question 11

Inferotemporal Cortex

Answer 11

-respond to shapes (combo of contours)
-FFA is here
-increase receptive field (entire visual field)
increase complexity
-Includes a number of subdivisions/channels/modules in which neurons show greatest responses to particular classes of objects (FFA)
-modular orginization (however activation is also distributed throughout)

Question 12

Heirarchical Coding

Answer 12

• Area lower down (like V2) location more important [NB for representation of the world]
• info hierarchical in ITC
• dif aspects of stim processed by dif anatomical areas but all must work together
• top more vauge (animate vs inanimate) snd lower is finer processing like (FFA or ob?), even finer (bottom would be is it eyes/lips of the face)