Vision • Photoreceptor cells • Rod & Cone cells • Bipolar Cells • Connect in betwee • Ganglion Cells • Go to the brain
Vision
• Photoreceptor cells• Rod & Cone cells
• Bipolar Cells• Connect in between
• Ganglion Cells• Go to the brain
VisionLGN
V1
V1
LGNganglion
cells optic chiasm: where theganglion cells cross so theleft side of each eye goes
to the right side of the
brain, and vice versa.
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
(broad tuning)
500 nm light
cone firing
blue green red
The Eye
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
(broad tuning) cone firing
blue green red
The Eye
b
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
b
b
b
The Eye
bipolar & horizontal cells
hh
hh
hh-
--
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
The Eye
bipolar & horizontal cells
• lateral inhibition
+ + + + +- - - - - - - -
Lateral Inhibition
9999111119999
00
+-
-
Lateral Inhibition
9999111119999
004+
-
-
Lateral Inhibition
9999111119999
004-4+
-
-
Lateral Inhibition
9999111119999
004-40+
-
-
Lateral Inhibition
9999111119999
004-4000-4400+
-
-
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
The Eye
bipolar & horizontal cells
• lateral inhibition• edge detection
ganglion cells
front view
Bipolar cells
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
The Eye
bipolar & horizontal cells
• lateral inhibition• edge detection
ganglion cells
receptive field
+-
cone cells:rod cells:
• periphery• movement• black and white
• fovea (center)• detail• color
The Eye
bipolar & horizontal cells
• lateral inhibition• edge detection
ganglion cells
receptive field
+-• center/surround
Center-Surround(Blob detector)
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-Surround
+
-
time
light position
firi
ng f
requ
ency
Center-SurroundHow’s it done?
Difference of Gaussians (Mexican hat)
light position
Distributed Visual Representation
• Different cells respond to different properties, such as bars of light at different orientations (i.e. the simple cells in V1).
• Different areas of the brain are dedicated to processing form and location information (i.e. the “what” and “where” systems, in the temporal and parietal lobes, respectively)
How does your brain put it together again?
•“red” neurons•“blue” neurons•“square” neurons•“circle” neurons•“upper-right”•“lower-left”
Binding Problem
How do we know which featuregoes with which object?
•“red” neurons•“blue” neurons•“square” neurons•“circle” neurons•“upper-right”•“lower-left”
Binding Problem
How do we know which featuregoes with which object?
•“red” neurons•“blue” neurons•“square” neurons•“circle” neurons•“upper-right”•“lower-left”
Binding Problem
How do we know which featuregoes with which object?
The Computations of Human Vision
Visual system must calculateObject color (Mostly known)Object shape
Begin with edges (Known)Find blobs (Known)How edges/blobs combine to form objects
(Mostly Unknown)Object movement (Mostly known)
The Computational System of VisionObject Motion
Marr’s levels:Computational
Determine motionRepresentation & Algorithm
Mostly known (but not by us!)Physical Implementation
Neurons in the eye and brain
Hierarchical processing
Low-level processing
High-level processing Examples
Biological motion
Object motion
Optic flow
Retinal motion
Hierarchical processing
Low-level processing/Low complexity
High-level processing/High complexity
• Biological motion
• Object motion
• Optic flow
• Retinal motion
…Superior Temporal Sulcus (STS)
Medial Superior Temporal area (MST)Middle Temporal Area (MT/V5)
Primary visual cortex (V1)Lateral Geniculate Nucleus (LGN)
Retina
Limits of motion perception
We can’t perceive motion that is either too FAST or too SLOW.
Verticalposition
time
Upward motion
Downward motion
Apparent Motion• “Broken” motion
• Stimulus flashing at different positions at different times
Very fast
Very slow
Beta Movement
PhiMovement
Optimal
Somewhat fast
Fli
cker
Rat
e
Phi Movement/Motion
• A “pure sense” of motion without seeing the intermediate steps
• No link-up/fill-in
Beta Movement/Motion
• Perceptually linking up the frames
• Smooth motion
• Motion picture technology
The Correspondence Problem
?
The Correspondence Problem
• How do we figure out which frame-2 dot should match with each of the frame-1 dots?
?
Frame 1Frame 2
The Wagon-wheel illusionPerceived directionReal direction
t1 t2
Slow CW Medium CW Fast CWSlow CW Ambiguous Slow CCW
Reality:Perception:
The Aperture Problem• Following from the correspondence problem• When the line’s motion is viewed through an
aperture, how do we figure out the “correct” motion?
The Aperture Problem• There are “infinitely many” possibilities.
or or ……
The Aperture Problem: a solution
• It’s not a problem:if the line has texture, or
if the line has endings, or
if the line is not straight, or…
…as long as there’s a UNIQUE POINT!
The Barberpole Illusion
• Perceived motion direction is parallel to the orientation of the rectangle
• Can be explained by “Unique-point heuristic”– Unique points are assumed to be on the long
edge
Per
ceiv
ed d
irect
ion
Optic Flow
Optic Flow
• Background scene flows as our we move
• We process motion signals at different locations to understand the optic flow pattern
• Optic flow is usefulfor inferring thedirection of self-motion
Optic Flow• Most studied flow patterns
Translational- object movement, eye movement,
etc. Rotational (or circular)
- head movement, eye movement, etc. Radial (expansion/contraction)
- motion in depth, self-motion, etc.
– They can represent most of the optic flows we see– Computationally, rotational and radial flows are
more complex than translational ones
Retina Motion vs REAL Motion
• Motion constancy– try tracking your moving finger!
• Retinal motion is combined with eye-movement to generate motion percepts
Motion in depth
• Retina is flat Motion signals are only 2D
• How can we know when something is moving towards/away from us?
• Try moving your finger towards your nose
Motion in depth
• The brain combines motion signals from the two eyes to infer motion in depth
Left eye Right eye
Far
Close
Right eye: leftward+) Left eye: rightward Approaching
Right eye: rightward+) Left eye: leftward
Receding
Biological Motion
• We are very sensitive to biological motion
• An analogy: Face in object perception
• Appears to require– extremely complex computations– a special motion processing mechanism
Motion Blindness• Patient LM
– Certain brain areas damaged through stroke– Almost all cognitive functions were intact
except for MOTION perception– She reported
• What she saw when pouring coffee into a cup: appears frozen like a glacier, does not perceive the fluid rising, often spills or overflows it
• “When I'm looking at the car first it seems far away, but then when I want to cross the road suddenly the car is very near”
– YouTube: http://www.youtube.com/watch?v=B47Js1MtT4wTitle: Akinetopsia (4:01)(a reproduced documentary for a class project)