Abstract
When viewing a scene, people have to summarize redundant information in a complex scene to overcome the limited capacity of the visual system. One way of summarizing information is perceptual averaging. The ability to compute average information is accurate and efficient across many stages of visual processing. However, the exact mechanism of averaging is not well understood. Here, we propose a distributed attention model of mean size perception. The model encodes and represents sizes with both early and late noise to reflect noisy percepts. It incorporates the central limit theorem to reflect noise cancellation by averaging. Finally, it has a component of attention to modulate the amount of attention allocated to each to-be-averaged item. The model predicts that averaging performance should increase as the number of to-be-averaged items increases, because of the central limit theorem. The slope of these increments should decelerate, especially for large set-sizes, because of late noise. Finally, the effects of attention should be manifested more in small set-sizes because of the limited capacity of attention. We tested these predictions using a mean size discrimination task. Critically, we varied the number of to-be-averaged items from 1 to 32. The model explained the observed data almost perfectly (r2 = .99) and the results were consistent with the model's predictions qualitatively. Observers' precision of averaging significantly increased with set-sizes. Increments in precision were prominent in small set-sizes and decelerated in large set-sizes. Attention parameters, assuming equal weights on all items (distributed attention), explained the increased precision of averaging in small set-sizes. Late noise parameters explained the decelerated precision of averaging in large set-sizes. Furthermore, the model explains why some previous studies, but not others, found set-size effects on averaging. The model provides a theoretical framework to interpret behavioral data and allows us to understand the characteristics of ensemble perception.
Meeting abstract presented at VSS 2018