psy 5018h: math models hum behavior, prof. paul schrater, spring 2004 vision as optimal inference...
Post on 19-Dec-2015
216 views
TRANSCRIPT
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Vision as Optimal Inference• The problem of visual processing can be thought of as
computing a belief distribution• Conscious perception better thought as a decision
based on both beliefs and the utility of the choice.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Hierarchical Organization of
Visual Processing
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Circuit Diagram of Visual Cortex
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Motion Perception as Optimal Estimation
QuickTime™ and aVideo decompressorare needed to see this picture.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Local Translations
OpticFlow:(Gibson,1950)
Assigns local image velocities v(x,y,t)
Time ~100msecSpace ~1-10deg
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Measuring Local Image VelocityReasons for Measurement Optic Flow useful:
Heading direction and speed, structure from motion,etc. Efficient:
Efficient code for visual input due to self motion (Eckert & Watson, 1993)
How to measure? Look at the characteristics of the signal
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
X-T Slice of Translating Camera
x
t
x
yQuickTime™ and aVideo decompressorare needed to see this picture.QuickTime™ and aVideo decompressorare needed to see this picture.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
X-T Slice of Translating Camera
x
t
x
y
Local translation
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Early Visual Neurons (V1)
QuickTime™ and aVideo decompressor
are needed to see this picture.
Ringach et al (1997)
y
x y
x
t
x
QuickTime™ and aVideo decompressor
are needed to see this picture.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
What is Motion?
As Visual Input: Change in the spatial
distribution of light on the sensors.
Minimally, dI(x,y,t)/dt ≠ 0
As Perception: Inference about causes of
intensity change, e.g.I(x,y,t) vOBJ(x,y,z,t)
QuickTime™ and aVideo decompressor
are needed to see this picture.
QuickTime™ and aVideo decompressor
are needed to see this picture.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Motion Field: Movement of Projected points
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Basic Idea• 1) Estimate point motions• 2) use point motions to estimate camera/object motion• Problem: Motion of projected points not directly measurable.• -Movement of projected points creates displacements of image
patches -- Infer point motion from image patch motion– Matching across frames– Differential approach– Fourier/filtering methods
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Problem: Images contain many edges-- Aperture problem
Normal flow:Motion component in the direction of the edge
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Aperture Problem (Motion/Form Ambiguity)
Result: Early visual measurements are ambiguous w.r.t. motion.
QuickTime™ and aVideo decompressor
are needed to see this picture.
QuickTime™ and aVideo decompressor
are needed to see this picture.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Aperture Problem (Motion/Form Ambiguity)
However, both the motion and the form of the pattern are implicitly encoded across the population of V1 neurons.
QuickTime™ and aAnimation decompressor
are needed to see this picture.
Actual motion
QuickTime™ and aAnimation decompressor
are needed to see this picture.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Plaids
QuickTime™ and aGraphics decompressor
are needed to see this picture.
QuickTime™ and aGraphics decompressor
are needed to see this picture.
This pattern was created by super-imposing two drifting gratings, one moving downwards and the other moving leftwards.
Here are the two components displayed side-by-side.
Rigid motion
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004Find Least squares solution for multiple patches.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Motion processing as optimal inference• Slow & smooth: A Bayesian theory for the combination of local motion signals
in human vision, Weiss & Adelson (1998)
Figure from: Weiss & Adelson, 1998
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Modeling motion estimation
From: Weiss & Adelson, 1998
L(v)∝ e− w(r )(I xvx+ I yvy+ It)2 / 2σ 2
r∑Local likelihood:
Lr (v)→ p(I |θ) ∝ Lr(θ)r∏Global likelihood:
P(V )∝ e− (Dv)t (r)(Dv)(r )/ 2
r∑Prior:
P(V )→ P(θ)
Posterior: P(θ |I ) ∝ P(I |θ)P(θ)
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Figures from: Weiss & Adelson, 1998
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Figure from: Weiss & Adelson, 1998
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Figure from: Weiss & Adelson, 1998
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Figure from: Weiss & Adelson, 1998
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Lightness perception as optimal inference
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Illuminant
surface reflectances
( )λxyS
€
L λ( )
Surface normal NLight dir. L
€
Simple rendering model
I(x,y) = S xy λ( )r L λ( ) ⋅
r N (x,y)[ ]
I(x,y)
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
400 500 600 700
400 500 600 700 400 500 600 700 400 500 600 700
( L , M , S )
REFLECTANCE ILLUMINANT SIGNAL
CONES
?
X =
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Land & McCann’s lightness illusion
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Neural network filter explanation
.
Differentiate(twice) byconvolvingimage with
"Mexican hat filter"
Lateral inhibition
Integrate
Threshold small values
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Apparent surface shape affects
lightness perception
• Knill & Kersten (1991)
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2004
Inverse graphicssolution
What model ofmaterial reflectances, shape,and lightingfit the image data?
ImageLuminance
Reflectance
Shape
Illumination
ambient
point
Shape
Reflectance
Illumination
point
ambient
x
different "paint" same "paint"