Abstract
In visual masking (Breitmeyer & Ogmen, 2006), a target stimulus is presented in close spatiotemporal proximity to a mask stimulus. The mask reduces the target visibility for certain stimulus onset asynchronies. Traditional explanations of masking assume disruptions in the forward visual information flow, either by independent channels or by lateral inhibition. Recently, a class of special masking experiments with common onsets of target and mask has been modeled by feedback flow from higher visual areas (Enns & DiLollo, 2000). Building upon this, we introduce a general model for visual masking inspired by the idea of perception as Bayesian inference. What an observer sees in masking experiments depends on posterior probabilities of competing perceptual hypotheses that are permanently checked against visual inputs. Masking effects result from an interaction between hypothesis prediction and hypothesis certainty. Parallel to our model, we performed masking experiments with stimuli composed of intact or scrambled Chinese characters, presented to a group of Chinese subjects and German controls. For Chinese subjects, for whom we assume higher prior probabilities for intact characters, intact character masks were more effective, as predicted by our model. The core components of our model can be related to lower and higher visual areas and its dynamics indicate functional roles of the circuitry of the visual system. We suggest visual masking to be regarded as an indicator that Bayesian inference is a crucial mechanism already in very early perceptual processing.
BreitmeyerB. G.OgmenH. (2006). Visual Masking. New York: Oxford University Press.
EnnsJ. T.Di LolloV. (2000). What's new in visual masking? Trends in Cognitive Science, 4, 345–352.
visual masking, modeling, Bayesian inference.