Visual Masking
Ch.5 pp.164-185
The retino-cortical dynamics
(RECOD) model
Models of visual masking
 Neural-network models
 Hartline-Ratliff inhibitory network (Bridgeman)
 Rashevski-Landahl two-factor network (Weissstein)
 RECOD (Breitmeyer and Ögmen)
 Perceptual Retouch (Bachmann)
 Boundary Contour System (Francis)
 Evidence for Transient-Sustained channel approach
 Transient channel (coarse spatial scales, information about temporal change
in the stimulus)
 Sustained-channel (fine spatial scales, information on stimulus form)
Outline
 Breitmeyer and Ganz’s sustained-transient dual-channel
model
 The RECOD model
 Theoretical rationale
 Temporal multiplexing
 Basic architecture
 The mathematical basis
 Unlumping: contour and surface
 Localization and visibility
 Next week: Explanatory scope of the RECOD model
Breitmeyer and Ganz’s sustained transient
dual-channel model (1976).

Main assumptions :
1.
Both target and mask activate long-latency sustained as well as short-latency
transient channels.
2.
Within a channel, inhibition is realized via the center-surround antagonism of
receptive-field. This is intra-channel inhibition.
3.
Between the two channels there exists mutual and reciprocal inhibition, the
inter-channel inhibition.
4.
Masking occurs in three ways:



5.
Via intra-channel inhibition (particularly in the sustained channel)
Via inter-channel inhibition (partic. transient-on-sustained inhibition)
Via sharing of sustained or transient pathways by the neural activity generated by
target and mask when they are spatially overlapping (intra-channel integration).
Transient channels signal the location, presence, rapid changes over time;
sustained channels signal patterns (Brightness, contrast and contour of slowly
moving stimulus)
Breitmeyer and Ganz’s sustained transient
dual-channel model (1976).

Main assumptions :
1.
Both target and mask activate long-latency sustained as well as short-latency
transient channels.
2.
Within a channel, inhibition is realized via the center-surround antagonism of
receptive-field. This is intra-channel inhibition.
3.
Between the two channels there exists mutual and reciprocal inhibition, the
inter-channel inhibition.
4.
Masking occurs in three ways:



5.
Via intra-channel inhibition (particularly in the sustained channel)
Via inter-channel inhibition (partic. transient-on-sustained inhibition)
Via sharing of sustained or transient pathways by the neural activity generated by
target and mask when they are spatially overlapping (intra-channel integration).
Transient channels signal the location, presence, rapid changes over time;
sustained channels signal patterns (Brightness, contrast and contour of slowly
moving stimulus)
Breitmeyer and Ganz’s sustained transient
dual-channel model (1976).

Main assumptions :
1.
Both target and mask activate long-latency sustained as well as short-latency
transient channels.
2.
Within a channel, inhibition is realized via the center-surround antagonism of
receptive-field. This is intra-channel inhibition.
3.
Between the two channels there exists mutual and reciprocal inhibition, the
inter-channel inhibition.
4.
Masking occurs in three ways:



5.
Via intra-channel inhibition (particularly in the sustained channel)
Via inter-channel inhibition (partic. transient-on-sustained inhibition)
Via sharing of sustained or transient pathways by the neural activity generated by
target and mask when they are spatially overlapping (intra-channel integration).
Transient channels signal the location, presence, rapid changes over time;
sustained channels signal patterns (Brightness, contrast and contour of slowly
moving stimulus)
Breitmeyer and Ganz’s sustained transient
dual-channel model (1976).

Main assumptions :
1.
Both target and mask activate long-latency sustained as well as short-latency
transient channels.
2.
Within a channel, inhibition is realized via the center-surround antagonism of
receptive-field. This is intra-channel inhibition.
3.
Between the two channels there exists mutual and reciprocal inhibition, the
inter-channel inhibition.
4.
Masking occurs in three ways:



5.
Via intra-channel inhibition (particularly in the sustained channel)
Via inter-channel inhibition (partic. transient-on-sustained inhibition)
Via sharing of sustained or transient pathways by the neural activity generated by
target and mask when they are spatially overlapping (intra-channel integration).
Transient channels signal the location, presence, rapid changes over time;
sustained channels signal patterns (Brightness, contrast and contour of slowly
moving stimulus)
Breitmeyer and Ganz’s sustained transient
dual-channel model (1976).

Main assumptions :
1.
Both target and mask activate long-latency sustained as well as short-latency
transient channels.
2.
Within a channel, inhibition is realized via the center-surround antagonism of
receptive-field. This is intra-channel inhibition.
3.
Between the two channels there exists mutual and reciprocal inhibition, the
inter-channel inhibition.
4.
Masking occurs in three ways:



5.
Via intra-channel inhibition (particularly in the sustained channel)
Via inter-channel inhibition (partic. transient-on-sustained inhibition)
Via sharing of sustained or transient pathways by the neural activity generated by
target and mask when they are spatially overlapping (intra-channel integration).
Transient channels signal the location, presence, rapid changes over time;
sustained channels signal patterns (Brightness, contrast and contour of slowly
moving stimulus)
Breitmeyer and Ganz’s sustained-transient
dual-channel model (1976)
 Forward masking
 Inter-channel inhibition
 Intra-channel integration
(structure, noise) and
inhibition (paracontrast)
 Near synchrony
 Intra-channel integration and
inhibition (as before)
 Backward masking
 Inter-channel inhibition
 Intra-channel integration and
inhibition
Breimeyer and Ganz (1976)
The retino-cortical dynamics (RECOD) model
(Ögmen 1993)
 How to deal with feedback processes:
theoretical rationale behind the model
 Mathematical perspective: need to avoid
unstable behaviour
 Trade-off between stimulus read-out and
perceptual synthesis in a feedback
system
The retino-cortical dynamics (RECOD) model

A solution: temporal mutiplexing.

The dynamics of visual processes
unfolds in 3 phases.
1.
A feedforward-dominant phase.
Strong afferent signals travel to cortical
areas allowing read-out of input.
2.
A feeback-dominant phase. Afferent
signal decays and feedback signal
establishes perceptual synthesis.
3.
A reset phase is initiated when inputs
change. A fast transient inhibition of the
feedback signal allows dominance of
the new input.
Purushothaman et al. (1998)
The retino-cortical dynamics (RECOD) model

A solution: temporal mutiplexing.

The dynamics of visual processes
unfolds in 3 phases.
1.
A feedforward-dominant phase.
Strong afferent signals travel to cortical
areas allowing read-out of input.
2.
A feeback-dominant phase. Afferent
signal decays and feedback signal
establishes perceptual synthesis.
3.
A reset phase is initiated when inputs
change. A fast transient inhibition of the
feedback signal allows dominance of
the new input.
Purushothaman et al. (1998)
The retino-cortical dynamics (RECOD) model

A solution: temporal mutiplexing.

The dynamics of visual processes
unfolds in 3 phases.
1.
A feedforward-dominant phase.
Strong afferent signals travel to cortical
areas allowing read-out of input.
2.
A feeback-dominant phase. Afferent
signal decays and feedback signal
establishes perceptual synthesis.
3.
A reset phase is initiated when inputs
change. A fast transient inhibition of the
feedback signal allows dominance of
the new input.
Purushothaman et al. (1998)
RECOD model : the basic architecture
 The magnocellular / parvocellular
pathways are identified with the
transient/sustained channels.
 Two layers: retinal ganglion cells and
LGN+cortical cells
 Two channels: fast-phasic M cells (left)
and slower tonic P cells (right).
 Each channel possesses both
positive and negative connectivity
patterns.
 Intra-channel integration and inhibition
for both M and P pathways
 Inter-channel inhibition
RECOD model : the mathematical basis
 p is the activity variable for
the cortical P cells.
 The first term ensures the
exponential decay of the
signal.
p
RECOD model : the mathematical basis
 The first excitatory term is
the feedback signal.
 ~p2 for small p and is
linear for greater values.
p
RECOD model : the mathematical basis
 The second excitatory term
is the afferent parvocellular
signal.
  is the delay between
magno- and parvocellular
pathways.
p
RECOD model : the mathematical basis
 Feedback inhibition
p
RECOD model : the mathematical basis
 Afferent parvocellular
inhibition
p
RECOD model : the mathematical basis
 Inter-channel transient-onsustained inhibition
p
RECOD model : contour and surface
 Example of model unlumping: contour
and surface dynamics
 The P pathway post-retinal network is
devided in two networks.
 Contour processing
 Surface processing
 A subcortical network is
added to account for
facilitatory effects in
paracontrast
RECOD model : contour and surface
 Metacontrast
 SOA of optimal suppression is
shorter for contour visibility
than for brightness visibility.
RECOD model : contour and surface
 Example of model unlumping: contour
and surface dynamics
 The P pathway post-retinal network is
devided in two networks.
 Contour processing
 Surface processing
 A subcortical network is
added to account for
facilitatory effects in
paracontrast.
RECOD model : contour and surface
 Paracontrast
 Maximal facilitatory effects on
contour visibility are found at
larger SOA than for brightness.
Explanatory scope of RECOD model :
localization and visibility
 Dissociation between target
visibility and target
localization in metacontrast
Next week...
 We will look closer at the explanatory scope of the
RECOD model.
 We will compare model simulations with results of
psychophysical experiments.