1. Why crowding?

-- Crowding is a big problem in amblyopia, affecting many aspects of their daily lives including reading. Thus, investigating whether it is possible to relieve crowding, as well as how it is done, is valuable.

2. Crowding seems to occur because of: excessive feature integration, improperly scaled attentional resolution (or attentional resolution limits) and location uncertainty about targets and features.

3. Since amblyopes generally have higher than normal positional uncertainty in the fovea, it follows that this would contribute to crowding in the amblyopic fovea.



Experiment 1: Does attention relieve crowding (more specifically, does shrinking the focus of the attentional spotlight to surround a target specifically and exclude distractors relieve crowding)?

-- no cue vs. precue vs. simultaneous cue (standardized across appropriate crowded and uncrowded acuities)

Experiment 2: Does location uncertainty about the location of the focus of attention affect the ability of attention to relieve crowding?

-- simultaneous cue with positional uncertainty (use acuity standardized to simultaneous attentional cue)

Critical measure in both experiments is the change in performance. Ultimate goal is to

Why do this first (i.e. before using more classic Feature Integration paradigms and models to study Feature Integration in amblyopia)?

-- neither of these questions have been studied in amblyopia; they have in normals. This provides a basis for comparison of amblyopic crowding and normal peripheral crowding and how they can be manipulated/explained with location uncertainty and spatial attention. It also allows for comparison of the shortcomings of these explanations between amblyopia and normal periphery.

THEN: let go of crowding and study feature integration in amblyopia.

In knowing how feature integration works or fails in amblyopia in comparison to normals, as well as knowing about how positional uncertainty and attention work/fail in relationship to crowding in amblyopia compared to normals, I can start to make actual predictions about the role feature integration plays in this whole mess.

-- future ability to make predictions based on amblyopic behavior vs. normal behavior re: attention, location uncertainty AND eventually more classic feature integration allow better prediction of how

To do: Program experiment. The Gaussian blob thing will be a challenge, but doable. Also the staircases.
Recruit subjects and get data. Expect approximately 8 hrs of subject time... 1 hr for acuities, 1 hr no cue, 1 hr precue, 1 hr simultaneous cue, 3 hrs simultaneous cue with uncertainty
Schedule more meetings. Hope experiment isn't ripped to shreds. Read Marisa Carrasco work on attention and acuity/spatial resolution.
Stay sane. ;-)

More reading areas:

Foveal vs. Peripheral Spatial Vision
-- Levi & Klein
-- Banks et. al. (1991) -- JOSA)
-- Anderson et al. (1991) -- J Physiology
-- Ferree et al (1931) * for optics of the periphery
-- Virsu & Rovamo (1979)
-- Levi et al (1985)
-- Rovamo et al (1982)

Phase Discrimination
-- Bennett & Banks (1987, 1991)
-- Field & Nachmias (1984)
-- Morrone & Burr (1988)

Task Difficulty in Visual Search
something by Jeremy Wolfe, John Palmer and/or Bill Geisler

Science is so going to eat my life. I will become one of those people who truly can't go more than three sentences without talking about what I study.

Textural analysis: visual computation of a statistic based upon an ensemble of objects.

Precision of estimates of information about individual features (at least in the case of orientation for oriented lines is independent of the number of features, which distinguishes crowding from other "masking" processes that seem to cause irretrievable loss of information.

Crowding = "elevation of target tilt threshold caused by (untilted) distractors that occurs even when there is no uncertainty regarding which objects are tilted and which are not."

Two explanations of crowding: Masking (distractors corrupt visual system's estimate of a particular kind of information about a target) Pooling (estimate is accessible only after it has been combined w/estimates of information about scene as a whole)

Task: detect orientation (cw/ccw) of "target" patch among distractors. 'Phase coherent' (targets are all windows on an underlying grating vs. 'Phase incoherent' (targets are of mutually independent phases)

Q: was array size matched w/V1 RF size? (if pooling is to be done within a single RF)

Generally threshold falls w/increase in number of targets (both with and without distractors). Early noise model does not explain what happens w/o distractors.

Pooling vs. Max SDT (latter assumes that observers base their responses solely upon the patch having the greatest apparent tilt assuming no interference from distractors). Max SDT works pretty well in no-distractor condition, but doesn't fit distractor condition.

Patch localization: Observer has to report orientation of targets (cw/ccw) as well as alignment of 3 targets. Reported alignment didn't improve with level of target tilt, suggesting access to a pooled signal only.

Q: what about target collinearity? e.g. would alignment be easier to see if, e.g. orientation of targets is 45deg CW and alignment is NW-SE

Fig 4. Cum gaussian has steeper slope when target has no distractors or distractors have similar orientation as target. Orthogonal distractors shift average tilt to near zero accross the display and flatten the slope of the psychometric function to the point that performance is essentially confounded.

Discussion: Crowding is generally viewed as undesirable, but it is also accompanied by an unimpaired ability of the observer to report a statistical property of a large field. So perhaps crowding refers to processing something that is not a texture as if it were a texture. How do we decide what constitutes a texture, then? What do we do when we want to disambiguate individual textural elements without foveating? Esp. relevant to amblyopia.

Phase coherence seems to matter in conditions of collinearity, though this does not necessarily mean that there are specialized mechanisms for collinearity detection.

Pooling may operate over more widely spaced elements provided that an observer does not know target location. This, however, can be relieved with attention. (Apparently not likely to happen in crowding). When a target does not "pop out, " visual system averages local estimates. Interesting to know whether this averaging occurs when there is a large difference between features that creates a pop-out effect.

Suggests lack of conscious access to V1 activity. (But what about top-down feedback?)