Purchase this article with an account.
Mikhail Katkov, Misha Tsodyks, Dov Sagi; The human contrast response function: overcoming experimental pitfalls. Journal of Vision 2007;7(9):261. doi: 10.1167/7.9.261.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
There were several attempts to characterize the human contrast-response-function (CRF) and the corresponding internal noise amplitudes (NA), but there is no commonly accepted solution. Our theoretical analysis of the Two-Alternative Forced-Choice (2AFC) task, using Signal Detection Theory (SDT), identified several important cases where an experimental solution is impractical due to the large number of trials required. Here we estimate the human CRF and NA for a Gabor stimulus (0–60% Michelson contrast of carrier grating), using the Category Rating Task and assuming a simple SDT model. Rating results showed large variations between experimental sessions which could not be explained by sampling error only. Nevertheless, some properties derived from the session data, such as CRF and NA were relatively stable. Threshold vs. Contrast (TvC) curves, computed from CRF and NA, generally agreed with the independently measured TvC curves using 2AFC methods and were similar across sessions and observers. In experiments with stable (across sessions) estimates, CRF and NA could be well described as having two regimes: (1) below the detection threshold - NA were decreasing function of contrast (in some sessions NA were constant or slightly increasing with contrast), with CRF having a high gain; (2) above the detection threshold - NA was relatively constant, and CRF had low gain with practically constant slope. We show that such CRF and NA can successfully fit data obtained from 2AFC experiments, including previous results described in Kontsevich, Chen and Tyler (Vis Res, 2002). Estimated NA is compatible with some models that assume stimulus uncertainty, but not with models that assume a maximum response decision-rule (Pelli, JOSA, 1985). It is possible that, in cases of uncertainty, decision is based on the summed activity within a population of detectors having properties defined by the uncertainty range, with uncertainty range decreasing with increasing input level.
This PDF is available to Subscribers Only