Summary Task 4 - from visual shapes to objects

4 – FROM VISUAL SHAPES TO OBJECTS
THEORIES OF OBJECT RECOGNITION

• Low level (early) vision à extracting basic features from the image (ex. Dots in the retina and lines
in the V1)
• Mid level (middle) vision à a stage of visual processing that comes after early vision and before
object recognition and scene understanding (high level vision)

• Perceptual organization à the process by which elements in a person’s visual fields become
perceptually grouped and segregated to create a perception
o Grouping à the process by which elements in a visual scene are grouped into coherent
units or objects
o Segregation à the process of
separating one area or object
from another

• Recognition-by-components (RBC)
theory à objects consisted of geometric
components called geons. We recognize
objects on the basis of the arrangement of
these geons
o Geons are a set of basic
geometric objects that are sufficient to constitute any object
o The visual system does this regardless of how the geons are oriented in space
o Human object recognition is not completely view-point independent
o A set of geons that would work to assemble all objects have not been proposed yet

,FROM SIMPLE LINES AND EDGES TO PROPERTIES OF OBJECTS

• RECALL:
o Cells in V1 are interested in basic features of a visual
image; respond to edges/lines of specific
orientations, motion, sizes etc
o Small and precise receptive fields (cell will respond
to the stimulus only if the stimulus is presented in a
very specific location relative to the point where the
person is fixating their gaze)
• Neurons in the striate cortex are activated by simple stimuli
and only respond if their preferred stimuli are presented in
very restricted portions of the visual field
• Object recognition requires a great deal of subsequent
processing and a large number of distinct visual processing
areas

Diagram: visual information enters at the bottom,
via the retinal ganglion cells (RGC), feed-forward
info flows upward. Visual processing is a two-
way street. From the LGN up, there are both feed-
forward & feed-back connections between areas

• Border ownership = when an object is sitting on a background, the edges defining the border
between the object and background belong to the object




• In area V2 many cells care about border-ownership
• In V2, a border ownership cell, which prefers a dark edge on the right side of its receptive field, would
respond more strongly to B, the edge of a black square, than to C, the logically identical edge of a
gray square, even though the receptive field is identically stimulated.

, OBJECT RECOGNITION IN THE BRAIN

• Extrastriate cortex = a set of visual areas outside V1 à V2, V3, V4 &
V5
o From V1, the dorsal (where/how) pathway goes to the
parietal lobe. It plays a role in the processing of object
location and generating actions to interact with objects
o The ventral (what) pathway goes to the temporal lobe and is
important for object recognition



• Traveling the what pathway from V1 into the temporal lobe, we can
see a progressive change in the responses of cells in different visual
areas
o V1 à cells respond best to lines and edges in very specific areas of the visual field
o V2 à sensitivity to ‘border ownership’ and illusory contours. Features to objects
o V4 à more complex attributes



• In V4, cells respond to more
complex attributes, but the perfect
set of stimuli for them has not been
determined yet

à in the first figure to the right:
darker circles indicate stronger
response from a V4 cell to the
shape in the circle. Apparently, this
cell responds to stimuli that point
to the right



à this cell responds to the
stimulus in (A) but not to the
stimulus in (B) because the black
object in (B) does not have a right-
pointing feature – its point is an
accident of occlusion by the red
object




Things get more complicated as we move farther into the temporal lobe