Abstract
Recent work using functional magnetic resonance imaging (fMRI) observed stronger blood-oxygen-level-dependent (BOLD) signals in visual cortex when observers were shown curved band-passed contours (snakes) compared to straight and parallel band-passed contours (gratings). To account for these patterns of neural activation, Fang et al (2021) propose a divisive normalization model in which the responses of visual cortical neurons are scaled by the variance in contrast across orientation channels, rather than by the local stimulus contrast. These fMRI findings and the proposed normalization model both predict that snakes and gratings should differ perceptually, and that these differences should be strongest at suprathreshold contrasts. To test these predictions, we performed psychophysical experiments in human participants with suprathreshold stimuli (n=5, 2300 trials). Observers reported which of two briefly flashed stimuli had higher contrast (contrast discrimination), which of two superimposed patterns appeared to be in front (monocular rivalry), and the time it took for a fixated pattern to begin fading (perceptual fading). Contrast discrimination thresholds and monocular rivalry were both biased toward snakes compared to gratings (p<0.001). Further, perceptual fading of these stimuli was proportional to the strength of fMRI (r=0.82, p<0.01) and model (r=0.91, p<0.001) responses. Specifically, during prolonged fixation, gratings faded faster than snakes from perception, and these differences were strongest at high contrasts and high densities. Taken together, these results demonstrate a new class of perceptual biases, validate the normalization model proposed by Fang et al (2021), and link BOLD signals to perception.