Rules by which the brain segments an object from the background: Evaluation of the Gabor model of simple cell receptive fields Beth Tuck Hanover College.
Post on 22-Dec-2015
215 Views
Preview:
Transcript
Rules by which the brain Rules by which the brain segments an object from the segments an object from the background: Evaluation of the background: Evaluation of the
Gabor model of simple cell Gabor model of simple cell receptive fieldsreceptive fields
Beth TuckBeth TuckHanover CollegeHanover College
4/13/20074/13/2007
BackgroundBackground
Brain must “make sense” of massive Brain must “make sense” of massive amounts of visual information to amounts of visual information to generate holistic picture of an objectgenerate holistic picture of an object
e.g.e.g.
Task of VisionTask of Vision
Task of VisionTask of Vision
Task of VisionTask of Vision
Task of VisionTask of Vision
Task of VisionTask of Vision
Figure-Ground PerceptionFigure-Ground Perception
World is divided into (1) figure being World is divided into (1) figure being inspected and (2) backgroundinspected and (2) background
SegmentationSegmentation
The process of separating a The process of separating a figure/object from background figure/object from background (Marr, (Marr, 1982) 1982)
Outline needed to segmentOutline needed to segment
Image from Marr (1982)
Simple Cell Receptive FieldsSimple Cell Receptive Fields
Simple cells (orientation specific) Simple cells (orientation specific) may assist in segmentation processmay assist in segmentation process
Krantz (1994). Sensation & Perception Receptive Field Tutorial : http://psych.hanover.edu/krantz/receptive/
Research QuestionResearch Question
How do simple cells (modeled under How do simple cells (modeled under the Gabor function) allow the brain to the Gabor function) allow the brain to segment visual information?segment visual information?– What are the “Rules” by which modeled What are the “Rules” by which modeled
cells segment visual informationcells segment visual information
Step 1: Develop ModelStep 1: Develop Model
Feed stimulus into programFeed stimulus into program Program presents stimulus to cells @ Program presents stimulus to cells @
18 orientations (rotates by 1018 orientations (rotates by 10oo increments)increments)
Program records output @ each Program records output @ each orientationorientation
Orientation and Output of Orientation and Output of ModelModel
0
500
1000
1500
2000
2500
3000
3500
0 10 20 30 40 50 60 70 80 90
Angle
Mo
del
Ou
tpu
t (P
uta
tive F
irin
g R
ate
)
Step 2: Bar Stimulus TestStep 2: Bar Stimulus Test
Feed bars of various widthsFeed bars of various widths Attempt to determine rules by Attempt to determine rules by
which these model cells allow which these model cells allow segmentationsegmentation
Example of Generation of Example of Generation of OutputOutput
Rule 1: Orientation of Model Cell Rule 1: Orientation of Model Cell with Greatest (Positive or with Greatest (Positive or
Negative) OutputNegative) Output At each location: find modeled cell At each location: find modeled cell
with maximum outputwith maximum output Plot orientation of that sensitivityPlot orientation of that sensitivity
– Most positive – draw in white Most positive – draw in white – Most negative – draw in red Most negative – draw in red
Most Positive
Most Negative
Rule 2: Greatest Output & Rule 2: Greatest Output & Limited Range of ResponseLimited Range of Response
At many locations, the responses of At many locations, the responses of all orientations are both:all orientations are both:– Relatively strongRelatively strong– All approximately the same valueAll approximately the same value
Limited Range of Model Limited Range of Model OutputOutput
In this case, model In this case, model cells don’t seem to cells don’t seem to indicate orientationindicate orientation
The square at the The square at the location is filled location is filled with an intensity to with an intensity to match the mean match the mean outputoutput
White for positive; White for positive; red for negativered for negative
0
500
1000
1500
2000
2500
3000
3500
0 10 20 30 40 50 60 70 80 90
Angle
Mo
del
Ou
tpu
t (P
uta
tive
Fir
ing
Rat
e)
Most Positive
Most Negative
+ Limited Range Activity
Rule 2 Close-upRule 2 Close-up
So far each location processed So far each location processed independent of surrounding locationsindependent of surrounding locations
What if output also depends upon What if output also depends upon adjacent responses?adjacent responses?
E.g. color where color depends not just E.g. color where color depends not just on response of one cone but all threeon response of one cone but all three– Red alone gives redRed alone gives red– Red and green gives yellowRed and green gives yellow
Rule 3: Local MaximumRule 3: Local Maximum
12
3
S1
S2
S3
0
2
4
6
8
10
12
14
16
18
20
Model Output (Putative Firing Rate)
X Position
Y Position
Plot Orientation of this Maximum
Rule 3: Local MinimumRule 3: Local Minimum
12
3
S1
S2
S3
0
2
4
6
8
10
12
14
16
18
Model Output (Putative Firing Rate)
X Pos
Y Pos
Plot Orientation of this Minimum
Local Maxima
Local Minima
+ Limited Range Activity
ConclusionsConclusions
Local Maxima/Minima + Limited Local Maxima/Minima + Limited Range seems to most accurately Range seems to most accurately recreate stimulusrecreate stimulus
Hermann-Hering gridHermann-Hering grid– Fovea – does not segment circleFovea – does not segment circle– Periphery – segments circle Periphery – segments circle
Verify with human subjectsVerify with human subjects
top related