Abstract
In metacontrast, a target is rendered invisible by a surrounding mask that appears after the target's offset, providing a tool for investigating temporal aspects of visual coding. Previous manipulations of target and mask duration have used constant stimulus intensity, so that longer-lasting stimuli appear brighter. When brightness is controlled by compensating longer duration with lower intensity, increasing target duration has little effect on U-shaped metacontrast, but increasing mask duration monotonically decreases target visibility when the mask follows a 10 msec target immediately. These results were simulated with four models: efficient masking (Francis), decaying-trace (Anbar & Anbar), two-channel (Weisstein, 1968), and lateral inhibition (Bridgeman). Using parameters in each model that had successfully simulated metacontrast in the past, the efficient masking model yielded a monotonically increasing function of visibility in as target duration increased, while the other models gave U-shaped functions. Testing the remaining models in the increasing mask duration experiment, the decaying-trace model gave an increasing function where the psychophysics showed a decreasing function, the two-channel model showed a plateau of high visibility at low mask duration while the psychophysical function began declining immediately, and the lateral inhibition model accurately simulated the results. We also tested a recurrent processing model (DiLollo, Enns & Rensink), which did well here but does not have the parameters to test the target-duration condition. The lateral inhibition model performed best overall, perhaps because it relies on distributed coding rather than discrete detectors to simulate target visibility. In a single layer of simulated neurons, each cell inhibits six near neighbors; secondary interactions spread stimulus-specific activity across the network.