University of Pennsylvania
Suboptimal visual averaging reveals compulsory nonlinear mechanisms in human vision
How humans integrate ambiguous and conflicting signals has been a focus of perception and decision-making research for decades. Identifying the source of suboptimality can reveal the information integration strategy used by the nervous system to solve ecologically relevant tasks in natural environments. In this study, we examined visual spatial averaging, a fundamental process underlying the transformation of noisy local signals into a more stable global estimate. A novel paradigm was used to measure the degree of human suboptimality against the ideal observer. In this talk, first I will establish that human visual averaging is suboptimal in both luminance and stereoscopic domains: the visual system does not faithfully compute the average of a spatially varied stimulus, especially when spatial variability is large. Then, I will present a series of experiments designed to identify the source of suboptimality. Our results indicate two kinds of nonlinear mechanisms at work: one is the down-weighting of outlier samples and the other is the over-weighting of spatially correlated samples. These compulsory nonlinear mechanisms, while suboptimal in our laboratory task, may reflect an information integration strategy that improves performance in more natural contexts cluttered with object boundaries and illumination variation.
A pizza lunch will be served.