Abstract
This study investigated how human observers perform mean brightness comparison between two briefly-presented stimulus arrays composed of different numbers of achromatic patches. With these stimuli, the array composed of more patches could contain the brightest patch even when its mean brightness was lower than the other’s. In Experiment 1, we asked observers to choose the brighter array on a dark background and found a mild, but consistent, bias favoring the brightest patch in the display. The array containing the brightest patch was chosen as brighter, even when its actual mean brightness was lower. However, the observed bias was smaller than the one predicted under the assumption that observers based their judgments only on the brightest patch. This suggests that the brightest patch received higher weight when its luminance was integrated with others. In Experiment 2, we asked observers to do the same task on a white background. The direction of the bias remained the same, although changing the background considerably altered apparent contrast of the stimulus and the brightest patch now exhibited the lowest contrast to the background (and thus was least salient). In Experiment 3, we changed the task and asked observers to choose the darker array. Then, the direction of the bias was reversed. Nonlinear brightness metric cannot explain the results, because the stimulus was completely the same as in Experiment 2. These findings implicated task-dependent weighted averaging; i.e., mean brightness comparison can be accomplished with flexibly relying more on a few items relevant to the task.