A biologically inspired psychometric function for accuracy of visual identification as a function of exposure duration

The psychometric function of letter identification is typically described as a function of stimulus intensity. However, the effect of stimulus exposure duration on letter identification remains poorly described. This is surprising because the effect of exposure duration has played a central role in modelling human performance in whole and partial report tasks in which multiple simultaneously presented letters are to be reported (Shibuya & Bundesen, 1988). Therefore, we investigated visual letter identification as a function of exposure duration. On each trial, a single randomly chosen letter (A-Z) was presented at the centre of the screen. Exposure duration was varied from 5 to 210 milliseconds. The letter was followed by a pattern mask. Three subjects each completed 54,080 trials in a 26-Alternative Forced Choice procedure. We compared the exponential, the gamma and the Weibull psychometric functions, all of these having a temporal offset included, as well as the ex-Gaussian, and finally a new psychometric function, motivated from single-neuron studies by (Albrecht, Geisler, Frazor & Crane, 2002). The new psychometric function stands out by having a nonmonotonous hazard rate which is initially rising from zero, then peaking, and finally decaying to a somewhat sustained plateau, mimicking closely observed instantaneous firing rates of monkey visual cortex neurons. The new psychometric function fits well to experimental data in both the present study and in a previous study of single-letter identification accuracy (Bundesen & Harms, 1999). Also, we conducted a follow-up experiment to test the ability of the psychometric functions to fit single-letter identification data, at different stimulus contrast levels; also in this experiment the new psychometric function prevailed. Further, after insertion into Bundesen’s Theory of Visual Attention (Bundesen, 1990), the new psychometric function enables closer fits to data from a previous whole and partial report experiment.