Download Kanizsa figures in current vision science

THE KANIZSA TRIANGLE AS A
CORNERSTONE OF VISION SCIENCE
JOHAN WAGEMANS
LABORATORY OF EXPERIMENTAL PSYCHOLOGY
UNIVERSITY OF LEUVEN, BELGIUM
KANIZSA LECTURE, TRIESTE, OCTOBER 29, 2010
Outline
• Part 1: A neurocomputational model for
the perception of Kanizsa figures
• Part 2: Kanizsa figures in current vision
science
PART 2
KANIZSA FIGURES IN CURRENT VISION
SCIENCE
Plan for Part 2
• Some key aspects of Kanizsa phenomena
• Some lessons to be drawn from them
• Some examples of on-going research (in
relation to Kanizsa project and beyond)
• Some reflections on the current status of
vision science
Some key aspects of Kanizsa phenomena
• Not contour completion and filling-in
• But surface construction based on 2-D
integration of difference signals
– local depth cues (L- and T-junctions)
– computation of border-ownership (edge
assignment) by interactions (multiple iterations)
– surface filling-in by integration
– depth-lightness linkage
• Depth perception = primary
• Differentiation-integration = general principle
Some general lessons to be drawn
• All perception is construction
– based on available information
– constrained by general principles (ecologically
valid)
– does not have to be cognitive or experiencebased (autonomous organizational processes)
• Surfaces are primary (Gibson, Nakayama,
…)
• Border-ownership is crucial
• Figure-ground segregation
BOWN
computation
Kanizsa
triangle
figure/ground
organization
depth
computation
3D reconstruction
More general lessons to be drawn
• Phenomenology is an important point of
departure
– discover what needs to be explained
• not necessarily a good guidance regarding
mechanisms
– end-point of experience may be misleading as
computational starting-point
• Deep and full understanding requires
– implemented model (all stages, from input image
to reproduced experience)
– based on general (not ad hoc) neurocomputational
principles (~inner and outer psychophysics)
– tested in wide range of conditions (Kanizsa
variation figures) (~experimental phenomenology)
phenomenology,
psychophysics
vision sciences
neurosciences
computational
modelling
Neural mechanisms of Kanizsa figures
• Some monkey single-cell neurophysiology
– von der Heydt et al. (1984)
– Zhou et al. (2000)
– Qiu & von der Heydt (2005)
• Some human fMRI
– Mendola et al. (1999)
V2 neurons with illusory contours
von der Heydt, R. et al. (1984). Science,
224(4654), 1260-1262
Neurons in V2 show responses to
illusory contour stimuli corresponding
to their preferred orientation.
Open question
Do these neurons represent “collinearity” (FACADE model) or
“BOWN as an implicit signal of surface” (DISC model)?
BOWN-sensitive
neurons exist in V2
Borderline-ownership sensitive neurones in V2
Zhou, H. et al. (2000). Journal of
Neuroscience, 20, 6594-6611
This neuron responds stronger
when the left white side is the
owner of the borderline.
BOWN-sensitive neurons in V2 are also
Borderline-ownership
sensitive
neurones
in V2
sensitive to
stereo
disparity
This neuron
responds
stronger
when the left
side is the
owner of the
borderline,
defined by
contrast (A,
C) or by
stereo
disparity (E,
F).
Qiu & von der Heydt (2005), Neuron, 47, 155-166
Illusory figures /depth perception
Mendola, J. D. et al. (1999) Journal of
Neuroscience, 19, 8560-8572
human brain activity for stereoscopic
depth and for illusory contours in
overlapping brain areas
Psychophysics of Kanizsa figures
• Brightness nulling
• Contour positioning
• Depth nulling
measuring brightness, contour and depth
perception
Previously used experimental techniques can now be used for
larger set of variation figures to test different aspects of the
model
brightness nulling
(Halpern, 1987, Spillman et al., 1984)
contour positioning
(Guttman & Kellman, 2004)
depth nulling
(Gregory & Harris, 1974,Whitmore et al., 1976)
stereoscopic presentation
L
R
adjusted gray scale (background=0)
brightness nulling experiment
figures
contour positioning experiment
distance from the contour (in pixels)
outside
0
inside
figures
depth perception
• adjustment task
depth nulling (on-going project)
Compare directly the perceived depth of different
Kanizsa figure variations
"in front“
“in-between”
“behind”
Application: Occlusion capture
• Perceived surfaces create new, strong
units (perceptual grouping, objectness)
Some examples of other work in our lab
• Overall ambitions:
– To study Gestalt phenomena and to try to
understand their underlying mechanisms by
making use of all tools and techniques from
modern vision science (multi-method approach)
– To (re-)integrate the understanding of Gestalt
phenomena into all that we know about vision and
the visual brain
• GestaltReVision program (funded by Flemish
Government, METH/08/02)
• www.gestaltrevision.be
Another application of the DISC approach
• Figure-ground multistability (perceptual
switching)
feedback influence to BOWN
feedback
figure side
+
ground side
random
numbers
perception of face or vase image
=
alternation of border-ownership
-+ +-
feedback enhancement of a perception
“It is a vase!”
“It is even clearer!”
feedback
feedback
-
+
-
++
multi-stable perception
DFV
300
face
0
vase
-300
1
200
iteration
400
600
800
1000
multi-stable perception
DFV
300
continuous presentation
face
0
vase
-300
DFV
intermittent presentation
300
face
0
vase
-300
1
200
iteration
400
600
800
1000
gamma/beta-prime distribution
disambiguation by Gestalt principle of “size”
-10
0 (control)
+10
+20
responses to disambiguated images
-10
F=875
V=125
0 (control)
F=480
V=520
+10
F=109
V=891
+20
F=0
V=1000
Some neuroimaging
• fMRI decoding of neural representations
– Jonas Kubilius et al. (submitted)
– Lee de Wit et al. (in preparation)
parts
corner
whole
11.4
8.4
-5.7
-4.1
V1
V2
V3
CS
CS
V1
V2
LO
V3
pFs
all > fixation
objects > scrambled
Diamonds are not forever
What happens to local orientation signals (V1) and motion signals
(V5) when percept changes from local to global?
Diamonds are not forever
What happens to local orientation signals (V1) and motion signals
(V5) when percept changes from local to global?
(real occlusion to disambiguate)
A lot of psychophysics
• On a wide variety of topics in mid-level
vision, incl.
–
–
–
–
–
–
–
perceptual grouping
figure-ground organization
adaptation aftereffects
masking
biological motion
haptic Gestalt formation
…
• Main focus: bridging gaps between levels
– low-, mid-, and high-level
– psychophysics, modeling and neural mechanisms
Patient studies
build-up paradigm, from random to perfectly
structured
– in children with autism (collaboration with Hull:
Tjeerd Jellema & Hollie Burnett)
• local – global
• natural – manmade
– in patients with brain deficits (collaboration with
Birmingham: Glyn Humphreys)
• different component processes: attention, grouping,
figure-ground, shape, identity, semantic, …
• neglect, agnosia, …
Visual perception and visual arts
• Announcement: conference in Brussels
(23-24 November 2010)
• Collaboration with contemporary, active,
interested artists
Anne-Mie Van Kerckhoven
• best-known artist
• works with a lot of different techniques (also
digital)
• important visual component
to her work:
perceptual organization
2D vs 3D
contours vs surfaces
figure-ground ambiguities
(e.g., perceptual holes)
– multiple depth layers
– depth cues
– …
–
–
–
–
Effects of art expertise
on individual scan paths
•
•
the artist
checking balance
Effects of art expertise
on individual scan paths
•
•
the novice
reading
Effects of art expertise
on individual scan paths
•
•
the expert
checking composition
Summary
• Kanizsa triangle has central place in
vision science
– crucial insights
– generic value (general lessons to be drawn)
– continuous source of inspiration
• vision science should
– be more interdisciplinary
– continue to take phenomenology and
psychophysics seriously
– try to relate more to other visual disciplines
(computer vision, arts)
THANK YOU!
MORE INFORMATION:
WWW.GESTALTREVISION.BE
[email protected]