Similarity grouping as feature-based attention.

Perception organizes spatially distinct areas of our world into groups, according to a set of well-studied cues, including connection, common region, proximity, and the Gestalt similarity cues of color, luminance, and shape. One counterintuitive recent proposal (Huang & Pashler, 2007; Levinthal & Franconeri, 2011) is that similarity grouping is 'just' feature selection - seeing a red, bright, or square group is just global selection of those features. This account makes the striking prediction that similarity grouping must be serial, such that other groups (green, dark, circles) cannot be constructed in the same instant. Past support - from serial grouping in display symmetry judgments and visual search tasks - has been subject to alternative interpretations. Here we provide the most direct evidence yet for this account, by demonstrating a striking lack of influence of similarity grouping in a task that is immune to these alternatives - when people rapidly estimate numerosity of a set of objects, grouping them reduces estimates. We replicate this effect (N=16) for connection (M=-27% reduction, p=0.001), common region (M=-17%, p=0.013), and proximity (M=-9%, p=0.033). But no significant reduction appears for color (M=2%), luminance (M=-5%), or shape (M=3%) grouping of element displays, despite a phenomenally strong grouping percept in these displays. The illusion of simultaneous grouping of similarity groups may be like the illusion that light in our fridge is always on - groups are always present when we select a feature, just like the light is always on when we open the fridge. But with a true test of simultaneity, similarity grouping can be shown to be serial. Meeting abstract presented at VSS 2015.