Purchase this article with an account.
Andrei Gorea, Jan delOho Balaguer, Vincent de Gardelle, Christopher Summerfield; Computing an average over space and time. Journal of Vision 2013;13(9):131. doi: 10.1167/13.9.131.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
How much time does it take to compute the summary statistics of an ensemble of stimuli? Bloch’s law stipulates that, within the temporal integration time for detection stimulus duration, T, can be traded for its intensity, I (or Contrast, C), while keeping performance constant. Granted linearity, Bloch’s law can be inferred from the C-sensitivity function of temporal frequency, TF (the Temporal Modulation Transfer Function, TMTF) of the system. Here we assess the visual system’s temporal integration characteristics for extracting one summary statistic of an ensemble of highly suprathreshold stimuli, specifically their average shape and average size. Using items whose shapes (from circles to squares), or sizes (circles of variable radii) were drawn over space or time from a normal or lognormal distribution, respectively, we measured subjects’ sensitivity in discriminating between the averages of two shape- or two size-distributions as a function of (i) the display time (27-800ms) of 2-to-12 circle-square shapes simultaneously displayed over space and (ii) the TF (1.2-38Hz) of 10 sequentially displayed circles at one single location. Both discrimination thresholds were assessed with a 2AFC+staircase procedure. The major two findings are that average-shape extraction is quasi-independent of presentation time and that average-size extraction is almost independent of TF (in both cases provided that the individual items remained visible). For the shape averaging experiment it is also found that performance is independent of the number of items (2-to-12) but decreases as expected with the standard deviation of the distribution from which they were drawn. Taken together the results point to the fact that, at least for the two dimensions studied, average extraction occurs at the detection threshold involving the readout of the same neural population code as for the extraction/identification of one single feature with a decision possibly based on the log likelihood function over the stimulated population.
Meeting abstract presented at VSS 2013
This PDF is available to Subscribers Only