Dear Marius

I am not quite sure from the description what task you are setting yourself with the discs. How can you demonstrate that the flattened globe has "a pretty certain location"?

With rectangular fields of dots side-by-side, the thresholds for depth discrimination behave very much as expected as a function of noise level. When Julie Harris and I tested this, it did seem as though the statistical integration of information across the field of dots was rather inefficient.

See EFFICIENCY OF STEREOPSIS IN RANDOM-DOT STEREOGRAMS

By:HARRIS, JM (HARRIS, JM); PARKER, AJ (PARKER, AJ)

JOURNAL OF THE OPTICAL SOCIETY OF AMERICA A-OPTICS IMAGE SCIENCE AND VISION

Volume: 9 Issue: 1 Pages: 14-24

DOI: 10.1364/JOSAA.9.000014

Published: JAN 1992

But, given your comments, you may be aware of this paper already.

An alternative approach is to apply a noisy variation in dot position to all the dots in the array but enforce the variation from trial to trial during the psychophysical task.

Kind regards

Andrew

Hi Andrew,

I wasn't aware of that paper per se, but there are some around that use similar methods. Let me explain my goal a bit better. What I'm trying to do is a 2-AFC task where people have to judge which of two stimuli (presented simultaneously) appears closer to them in depth. The stimuli are disc-shaped and consist of a large number of dots. I measure along a psychometric curve, and would expect the slope to decrease with increased disparity noise (or the JND to increase). But it doesn't seem to matter at all how much noise is added. My thought was that with hundreds of dots, the distribution of disparity is sampled so exhaustively that the observer has no problem estimating the centre of the distribution and this is what an ideal observer would use to respond in each trial. So with fewer dots it should work more like how I want it to work. However, when I do that the dots stop forming a disc that can be perceived as one coherent object with one depth assigned to it. When this works I need to add other stimuli, so this last point is kind of important.

I'll have a look at that paper and see if I can improve the paradigm. Thanks!

Marius

Another way of making the depth judgement harder would be to reduce binocular correlation, i.e to. replace a fraction of the dots in one eye with randomly positioned dots. However, It is possible that this is not what you want since the perceptual appearance of binocularly uncorrelated dots is quite different from that of a cloud of dots. See e.g. the paper "Independent Mechanisms for Bright and Dark Image Features in a Stereo Correspondence Task" (Read, Vaz & Serrano-Pedraza 2011, Journal of Vision) for a discussion of the two different methods and how they might differ in what stage of visual processing is affected by the manipulation

Frederik, I tried similar stimuli on myself and I didn't like them (admittedly a very imprecise criterion). What I guess is happening in them, is that they start to become binocular rivalry stimuli. This would mean that some dots have no depth at all, which may result in uncertainty about depth... or it may not. Thanks for the thought though!