Transcript ExtraStriate

VS131 Visual Neuroscience
Extra-Striate
Cortex
Hey stupid: remember to distribute the handouts this time!
Grossly simplified
block-diagram of
the cortical visual
system
Another grossly
simplified blockdiagram of the
cortical visual
system
Even more
grossly
simplified blockdiagram of the
cortical visual
system!
You will never
see the cortical
visual system
drawn the same
way twice. You
have been
warned!
Nota Bene:
Almost all of our current knowledge of extrastriate visual cortex comes
from work in the rhesus monkey. All current textbooks refer to monkey
data.
Recently, work using functional Magnetic Resonance Imaging (fMRI),
which unlike other research techniques can be applied in both humans
and primates, has started to tie the monkey data to the human data.
It appears that, indeed, much of the basic organization of the cortical
visual system in monkeys is the same as that of humans.
Work is ongoing and as usual there will doubtless be differences
between the species, especially in ‘higher’ (farther away from V1) visual
cortical areas.
Often you may read of the “human homolog of V4” instead of just V4.
Rhesus monkey cortical visual areas unfolded
Human visual cortical areas have about the same layout as
the rhesus monkey, but are only about 20% of cortical surface
A neurologists’ view of cerebral cortex (does not cover specific areas):
Occipital: VISION
Parietal: VISION + integrating sensory information different parts of the body
and motor control + stuff
Temporal: VISION + hearing + speech/language + memory + stuff
Frontal: Motor cortex, “executive function” (planning), + stuff + vision esp.
VOLUNTARY SACCADIC EYE MOVEMENTS.
Very crudely:
DORSAL PATHWAY: “WHERE” stream of visual processing: magno.
VENTRAL PATHWAY: “WHAT” stream of visual processing: parvo.
V2
-> Closely associated with V1
-> BOTH magno and parvo sections
-> Neurons in V2 have, at first glance, roughly similar
properties to those of neurons in V1. Small receptive
fields, often selective for orientation of a visual stimulus.
(Macaque)
Cytochrome
Oxidase stain
V1 blobs -> V2
this stripes
V1 ‘interblobs’
(I.e., not blobs)
-> V2 interstripes
V1 layer 4B ->
V2 thick stripes
Another view –
note ‘illusory’
border from
offset lines in D.
V4
-> One of the larger and more important of the post-V1/V2
visual cortical areas
-> More of a ‘parvo’ sort of pathway. ‘Ventral Stream’
-> More complex responsivities to shapes than just
oriented lines (real or imagined). Larger RF size.
-> Complex color properties (used to be thought of as the
color center, now know that it does more than just color)
-> Attentional effects become easier to elicit
-> Lesions cause more subtle problems than just visual
field defects.
V4 has larger RF’s – here we can fit two stimuli inside. This is from
experimental data from a monkey recording from a single V4 neuron. If
the monkey is attending to a location where an effective stimulus (I.e.,
one that elicits a strong response) is present, you get a strong response
(left). However, if the monkey is not paying attention to that location, you
get a much weaker response, even though the optical stimulus is
identical (right).
As before but now the monkey is only paying or not paying attention to
the entire receptive field. There are lots of different variations of this
basic paradigm.
Very much harder (though not impossible) to see such effects in V1/V2.
V4 neurons often have more complex shape selectivities
V4 neurons often have more
complex color properties
Our perception of color is
NOT a simple function of the
amount of red, green, and
blue light. (Ever take a
picture using daylight film
and incandescent light?).
A V4 neuron responding to a
red square in the middle of a
lot of different color patches
(‘Mondrian’).
It responds strongest when
the rest of the scene is lit by
white light (RGB), weakest
when everything is lit by red
(R) thus making color
distinctions impossible.
In experimental animals lesioning V4 does not result in specific visual
field defects, but in more generalized problems of shape perception. For
example, an animal with a V4 lesion may be able to learn to recognize
an object in one orientation, but be unable to recognize it if it is viewed at
a different angle or distance.
Also, lesions of V4 have been associated with problems in color vision.
IT
-> Inferior temporal cortex (IT, TE, TEO, etc.)
-> Last of the hierarchy of (more-or-less) purely visual
areas in the ventral (‘parvo’, ‘what’) stream.
-> Neurons can have very large receptive fields…
-> …but at the same time, the specificity for visual stimuli
can be very high
-> Lesions of IT can have devastating consequences for
the ability to recognize specific objects (e.g. faces) with no
corresponding loss of acuity or visual field deficits.
-> Large attentional effects
-> Lesions of temporal cortex can cause visual field
deficits by interrupting the passing fibers of Myer’s loops.
From Van
Essen and
Colleagues
RF sizes
get larger
further
along in the
looser
hierarchy of
visual
areas…
(And by IT
they often
cross the
midline)
IT neurons
in the same
cortical
column tend
to have
similar (not
identical)
properties
Remember the
columns in V1?
It has been
speculated that
there is a similar
modular
organization in IT,
but if so, we do not
know what the
analogs of location
in space,
orientation, or
ocular dominance
are.
“Grandmother Cell” Hypothesis
IT cells can be so specific in their response properties that it
seems like a neuron might only fire when one specific object
(like someone’s grandmother) comes into view. Thus, if the
neuron that is ‘tuned’ to your grandmother fires, that is a
signal to the rest of the brain that grandma is there.
However, while IT neurons have very specific response
properties, they still do respond to more than just one object.
Current thinking is that the presence of a particular object
(face) is coded for by a pattern of of activity of a relatively
small (but still greater than one) number of IT cells. Thus, at
least for now the ‘grandmother cell hypothesis’ is no longer
generally accepted.
MT/V5
-> Middle Temporal cortex (MT, but only really middle temporal
in owl monkeys!) related areas are Mst, and various parietal
something-or-other).
-> In the hierarchy of (more-or-less) purely visual areas in the
dorsal (‘magno’, ‘where’) stream.
-> Specialized for motion processing
-> MT is not large but it appears to be important and it has
been studied a lot, so it gets a lot of coverage in textbooks.
-> Lots of heavily myelinated axons, fast conduction velocities
(a magno trait). Stains heavily for myelin compared to other
visual cortical areas.
-> Cells in MT are really easy to activate with moving stimuli.
Analysis of Motion
As usual, what seems easy is really incredibly hard. Figuring
out what is going where in a visual scene is a very hard
computational task – one that we are not even close to solving
for computer vision. For starters, we sense motion when our
eyes are fixed and an image moves across the retina, and
equally when the eyes track a moving object and the image on
the retina is fixed.
The visual system can
also infer motion from
isolated sightings of an
object in different
locations at different
times: Apparent Motion.
The visual system can also solve the Aperture Problem.
Unlike neurons in V1, some (not all!) neurons in MT appear to solve the
aperture problem: they respond to the integrated motion of the entire object,
not just to the motion of its separate parts.
MST: Medial Superior Temporal area
-> Gets most of its inputs from MT
-> Neurons are tuned to flow fields.
Patterns of optic flow
The parietal lobe is in many ways the least well understood of
the cortical lobes. Contains primary somatosensory cortex, and a
lot of regions dealing with vision. Important for hand-eye
coordination and regulating attention, possibly creating a ‘sense
of space’. Lesions can cause hemi-neglect (“eyes right). Some
areas found in monkey posterior parietal cortex:
LIP: Lateral Intraparietal Cortex. Contains a retinotopic map of
salience. Neurons respond well to recent stimuli, or to stimuli
that are behaviorally relevant. Predictive remapping across
saccades!
VIP: Ventral Intraparietal Cortex: A multisensory map of space
around the mouth and face, important for ingestive behaviors?
Other cues to
relative size and
motion have to
be evaluated by
the visual
system, and
integrated with
all the other
sources of
information.
Binocular Disparity:
Judging distance by
triangulating between the
two eyes.
First neurons sensitive to
binocular disparity are
found in area V1. Disparity
sensitive neurons are found
throughout visual cortex,
but are most easily and
commonly found in the
dorsal/parietal pathway.
Near disparity = crossed
Far disparity = uncrossed
Single-Image Random-Dot Stereogram!
Autostereogram
Random-Dot Autosterogram
“Magic Eye”
So how do they work?
Standard stereograms use two images, one for each eye. Various
optical systems ensure that the left eye gets one image, and the right
eye gets another image.
For autostereograms, the bottom line is that there are two hidden
pictures in a single image seen by both eyes, the two hidden images
have different amounts of horizontal shift in different regions of the image
that are interpreted by the brain as different depths. Using random dots
hides the two images well, because as with standard random dot
stereograms, you can’t see the images unless they have been
binocularly fused.
Actually looking at
two objects, one
nearer and one
farther than the
fixation point.
The brain interprets
the shifts in the
image of the objects
as evidence of them
being nearer or
farther than
fixation.
Standard
stereogram, each
eye gets a separate
image.
As before, the brain
interprets the shifts
in the image of the
objects between the
two retinas as
evidence of them
being nearer or
farther than
fixation.
Single image
stereogram, two
shifted copies of
each object, the
brain again
interprets the shifts
in the image of the
objects between the
two retinas as
evidence of them
being nearer or
farther than
fixation.
BUT: How do you
hide the two
images?
Very crudely: use random-dot stereograms. You can’t see the
objects unless they are binocularly fused, thus hiding the two
pictures. At each place in the single image, you have two
overlaid (transparent or alternating) random dot stereograms
with different horizontal shifts. The brain uses the shift that
produces the best match to estiamte distance.
Why does it take time to see these images? Because in
normal conditions there are false binocular matches
everywhere, and the visual system ignores them unless there
is confirmatory data. Initially all the little random dot
stereogram patches are filtered out. Once a few manage to
break through at once and give the sense of a colinera
contour, this reinforces the depth sense from the random dot
stereograms and the entire picture snaps into focus.
Tricks for viewing:
Most autostereograms are designed for wall-eyed (diverging)
viewing. Need to focus (converge) on a point behind the image
plane.
People with stereo vision deficits can’t fuse random-dot
stereograms, including magic eye (between 1 and 5 percent of
population, give or take).
-> Use bright lighting: causes pupil to constrict, decreases
conflicting information due to blur and depth-of-focus.
-> Try to look behind the image. Look at your reflection if on
glossy paper (which optically will appear twice as far away). Hold
up to your nose and slowly pull away. Look at an object behind
the image plane (opposite if autostereogram designed for crossed
viewing).