Transcript Explanatory scope : Dual

Explanatory scope : Dual-channel
RECOD model
Chapter 5, Pages 186-218
Harsha KASI
PhD student, Institute of Microsystems and Microelectronics
EPFL
Remember
• Original sustained-transient model & RECOD model
share common mechanisms critical to masking
• Chapters 1, 2 and some additional results introduced in
this chapter
• Scope of 2 models by representative set of findings
17 July 2015
Explanatory scope : Dual-channel RECOD model
Outline
 Justification for effects and experimental
findings – comparison of model simulations and
psychophysical experiments
 Explanatory power
 Comparisons and critiques
17 July 2015
Explanatory scope : Dual-channel RECOD model
Paracontrast and metacontrast suppression
Flicker persisted longer in the middle band
Flicker persisted longer in the double white arcs
Critical flicker frequency (CFF) : former
latter ↓↑
Paracontrast suppression
Sherrington (1897)
(target)
(mask)
latter
former
(mask)
(target)
Metacontrast facilitation → Metacontrast suppression ?
Piéron (1935) – not only CFF but on brightness perception as well → metacontrast
suppression
 Metacontrast suppression (Brightness) – Faster transient activity by the
lagging flash inhibiting the slower sustained response of the leading flash
 Paracontrast suppression (CFF) – Slower sustained activity by the first stimulus
reciprocally inhibiting the faster transient (flicker) by the second stimulus
1st stimulus: higher CFF relative to the inhibited CFF of the 2nd stimulus
17 July 2015
Explanatory scope : Dual-channel RECOD model
Transient masking effects
30-ms sinusoidal target grating at on-and offset of
a 700-ms luminance flash mask (@ 54.8 cd/m2)
transient mask overshoots
Assume:
1.0 c/deg – low spatial frequency transient
7.8 c/deg – high spatial frequency sustained
Peripheral transient activity by mask flash adds
‘noise’ to the ‘signal’ of transient channels and not
sustained channels
1.0 c/deg: SNR or Weber ratio in transient channel ↓
→ overshoots!
7.8 c/deg: only a sustained masking effect at mask
onset or offset
Green (1981)
17 July 2015
Explanatory scope : Dual-channel RECOD model
Transient masking effects 2
Mitov et al. (1981): Overshoots inversely proportional to the spatial frequency of the
grating – 2 c/deg 6 c/deg 18 c/deg
Spatial frequencies at and below 6-c/deg, and with spatial frequency increase if the
magnitude of transient activation decreases and that of sustained channel
increases
Teller et al., Matthews (1971): No overshoots with mask sizes e.g. <30’. However,
with larger masks (e.g. > 60’)
Low spatial frequency gratings, optimal for activating transient channels under large
conditioning flash mask
Breitmeyer and Julesz (1975) and Tulunay-Keesey and Bennis (1979): overshoots
found in Green and Mittov’s studies depend on the rise and fall times of the mask at
its on- and offsets
Slowly ramped instead of abrupt on- and offsets attenuate the transient response
leading to curbing the transient masking overshoots
Matsumara’s (1976) work provides evidence to this !
17 July 2015
Explanatory scope : Dual-channel RECOD model
Contour and Surface dynamics – Unlumped p-pathway
Metacontrast: Strongest at shorter SOA
for the contour compared to the
surface/brightness network (20 ms vs. 60
ms)
Paracontrast:
Contour network – a long-lasting
suppression coupled with a strong
suppression – SOA ~ -10 ms
Surface network – Weaker long-lasting
suppression and then enhancement
Identical set of equations with different
weightings associated with inhibitory
and facilitatory processes
17 July 2015
Explanatory scope : Dual-channel RECOD model
M/T Ratio – type B to A metacontrast
Transition : mask/target energy ratio is greater than unity
Difference in masking contrast thresholds
Type B @ lower mask contrasts and
produced by a high-gain, rapid-saturation
transient-on-sustained inhibition
transforms to a
Type A @ high mask contrasts, produced
by a low-gain linear intra-channel
sustained-on-sustained inhibition
superimposed on the former inter-channel
inhibition
17 July 2015
Explanatory scope : Dual-channel RECOD model
Dichoptic type A forward and backward masking
Dichoptic type A forward masking by noise or structure is
typically weaker than type A backward masking
(Greenspoon and Eriksen 1968, Turvey 1973)
Forward masking by structure or
noise:
Post-retinal transient activity can
locally inhibit post-retinal sustained
mask activity → less masking by
integration
On the contrary, backward masking:
Sustained mask activity intrudes
unobstructed into target’s sustained
channels and @ post-retinal levels
transient
mask
activity
inhibits
sustained target activity → facilitate
intrusion – more masking !
Since these interactions are dichoptic,
very likely exist at cortical levels
17 July 2015
Explanatory scope : Dual-channel RECOD model
Left
eye
Right
eye
Monoptic – Type A forward masking stronger than type A backward masking
Integration of target and mask activities occur early – photoreceptor and postreceptor neural levels prior to the centrally located sustained-transient
inhibitory interactions
Type A forward and backward pattern masking as well as type B para- and
metacontrast are obtained dichoptically and monoptically (Alpern 1953;
Michaels and Turvey 1973, etc.)
Either integration in common sustained pathways or inter-channel inhibition
Type B metacontrast effects ↓ in magnitude as the spatial separation between
the target and mask stimuli ↑ (Alpern 1953; Breitmeyer and Horman 1981, etc.)
Spatially restricted receptive fields of sustained and transient neurons & the
topographical mapping between retina to the visual cortex (Brooks and Jung,
1973)
17 July 2015
Explanatory scope : Dual-channel RECOD model
Reaction times – contour interactions
In paracontrast, reaction times (∆RT)
for target localisation increase
Paracontrast: Both data and model
show an inverse U function
Metacontrast: constant function
Paracontrast: Both data and model
show an inverse U function
Metacontrast: constant function
Change ∆RT in reaction times due to contour interactions between the target and mask as a function of
SOA for two M/T contrast ratios. The middle curve corresponds to the average of the M/T=3 and
M/T=1 data. Error bars represent ±1 SE of the mean. The squares are the predictions of the model.
Reproduced from Ögmen et al. 2003
Paracontrast: close examination of M/T=3 and model – an inverse W
function; peaks and dips shifted w.r.t each other
2 peaks in the W-shaped function – separate contributions of inter-channel
sustained-on-transient inhibition and intra-channel transient-on-transient
inhibition to reduce activity of the transient channels responding target
17 July 2015
Explanatory scope : Dual-channel RECOD model
Werner (1935): Metacontrast masking of a target pattern is inversely related to
the orientation difference between target and mask stimuli
Cortical transient as well as sustained neurons are orientation selective (Ikeda
and Wright 1975; Stone and Dreher 1973)
Mutual inhibition between cortical orientation-selective cells is itself orientation
selective (Benvento et. al. 1972, etc.)
Blurred mask does not substantially reduce metacontrast of a non-blurred target
Transient channels are insensitive to high spatial frequencies and so to image
blur (Growney, 1976)
Single-transient paradigm (Breitmeyer and Rudd 1981): Brief mask suppresses
visibility of a prolonged sustained peripheral target for several seconds
Single-transient stimulus can activate transient-on-sustained inhibition, so
despite the necessary 2-transient paradigm in metacontrast, contrary to Matin
(1975).
Activation of T-M neurons is not required, transient neuron activation by mask
alone is sufficient
17 July 2015
Explanatory scope : Dual-channel RECOD model
Target recoverability
Addition of a second mask (M2) to a target (T) and primary metacontrast mask
(M1) can lead to the recovery of visibility of the target
Two effects:
1. M2-T-M1 :
Target visibility recovered
No change in the visibility of prim.
mask M1
2. T-M1-M2 :
No change in target visibility
A reduction in visibility of M1
Double disassociation, i.e., visibility and metacontrast masking effectiveness
associated with sustained and transient responses
Target recovery: M2 inhibits M1’s transient activity sustained-on-transient
inhibition
M1 reduced visibility : inter-channel transient-on-sustained inhibition by M2
17 July 2015
Explanatory scope : Dual-channel RECOD model
Comparisons and Critiques
Apart from explaining various effects (Chap. 2), 2 models give adequate
explanation of many variations of them as well
Revised version of Weisstein et al. 1975 accounts for metacontrast –
transient-on-sustained inhibition of the non-recurrent forward type
For paracontrast: sustained-on-transient inhibition of the non-recurrent forward
type
In conformance with assumption 1 of the Breitmeyer and Ganz’s model
Differs in assumptions 2 and 4, which in Breitmeyer’s models states that:
Paracontrast is realised via intra-channel inhibition effected in the sustained
channels, rather than Weisstein’ et. al’s corresponding assumption of interchannel, sustained-on-transition inhibition
Weisstein’s model cannot adequately account for the absence of type B
metacontrast when simple reaction time or detection rather than brightness
perception are used as criterion responses
17 July 2015
Explanatory scope : Dual-channel RECOD model
Comparisons and Critiques 2
Matin’s (1975) model with the sustained-transient model is not so similar in
regard to required activation of T-M neurons
T-M neurons → transient; T neurons → sustained
Shorter response latency of T-M neurons compared with T neurons bears a
similarity with the sustained-transient model
This combined with the inter-channel inhibition is equivalent to the
assumptions 1 and 3 of the sustained-transient model and the fast-inhibition
hypothesis of Weisstein’s
RECOD model
Converges to sustained-transient model
Incorporating recent neurophysiological findings, feedback mechanisms,
proposing additional feedforward, feedback-dominant phases of operation,
explicit network structure and a quantitative description that can be simulated
and compared directly with the experimental data
17 July 2015
Explanatory scope : Dual-channel RECOD model
Comparisons and Critiques 3
Feedback structure aspect of RECOD model makes it comparable to some
discussed in Chap. 4
Dual-channel aspect of RECOD model makes it significantly different from:
Bridgeman’s (1971, 1977, 1978) neural-network model
Ganz’s (1975) trace decay-lateral inhibition model
Reeves’s (1981) non-neural models
None of the neural or non-neural models incorporate the distinction between
transient response components and slow sustained ones which can
reciprocally inhibit each other
RECOD model incorporates:
Feedback (recurrent) connections as in Bridgeman’s single-channel model
Dual-channel structure to avoid spatiotemporal blurring that would occur in
Bridgeman’s model so that perceptual dynamics can be organised as entities
and can be processed individually
17 July 2015
Explanatory scope : Dual-channel RECOD model
Summary
A review of psychophysical studies of spatiotemporal properties of human
vision characterised by:
1. Separate pattern and movement or flicker thresholds
2. Temporal integration and persistence
3. Reaction time and effects of flicker adaptation
all as a function of spatial frequency → existence of sustained/transient
channels
Supported well by neurophysiological evidences – two parallel afferent
pathways with similar characteristics
RECOD model adequately accounts for a wide range of
visual masking phenomena discussed throughout this
book !
17 July 2015
Explanatory scope : Dual-channel RECOD model
Thank you
for your attention
17 July 2015
Explanatory scope : Dual-channel RECOD model