Predictive Coding Correctly Predicts Perceived Chromatic Negative Afterimages

Although the opponent model has traditionally been considered to explain the mechanisms underlying visual afterimages, recent studies suggest that some predictions of the opponent model are obviously incorrect. In the current study, we modeled afterimages using predictive coding, in which perception is generated via efficient information processing. The only assumption we made to apply information processing to the afterimage process was that " if the input information remains constant beyond the limit of brain memory, perception shifts toward a reasonable value (the average of past values)." In our experiments, because the inducer color did not change, its perception was expected to shift to gray due to adaptation. In contrast, because the reference color changed to the opposite color every second, its perception was expected to be maintained. Systematic comparisons of the model predictions with participants’ perception revealed high accuracy. Importantly, this model is the first method to continuously present accurate afterimage predictions on the same screen as the inducer at the same time. This showed that, contrary to previous reports, adaptation varied with inducer presentation time, and the afterimage changed accordingly.These results suggest that the brain constructs perception via information processing, and that visual afterimages result from its effects.