Abstract
In human rod vision, detection threshold for small, brief flashes is proportional to the square-root of adapting luminance over a considerable range. This finding was classically explained in terms of an ideal observer whose visual sensitivity is limited by photon noise from the adapting field, integrated over the flash area and duration (Rose, 1942; de Vries, 1943). Here, we employed a dichoptic brightness matching technique to study square-root adaptation in humans (Brown & Rudd, 1998). Flashes in one eye—whose adaptive state remained fixed—were matched in brightness to flashes presented simultaneously to the other eye, whose adaptive state was manipulated. A square-root brightness law held for superthreshold flashes over the same adapting range as the threshold law. Thus, the square-root law does not apply only to threshold, but rather characterizes the properties of a monocular gain control mechanism. We measured the time required for this mechanism to adapt half way to changes in mean luminance to be ~100 sec. Gain changes were elicited by adding random flicker to the adapting field while keeping the mean level fixed, implying adaptation to noise. Consistent with these observations, we quantitatively modeled visual threshold by assuming that the gain applied to the flash is set by a leaky integral (tau = 100 sec) of the photon noise summed over the adapting field area in which the flash is presented. Our results are consistent with a slow physiological adaptation to photon noise occurring over a duration >10,000 times that our 10 msec test flashes.