Abstract
Through the influence of grouping cues such as physical connection or sharing similar features, we often treat objects in the world as belonging together. While there is ample work exploring the behavior of these cues and their relative dominance, less is known about the mechanisms underlying how they are processed. Using a novel paradigm, we demonstrate a fundamental division between the mechanisms of connectivity grouping, which appears to be processed in parallel across the visual field, and similarity grouping, which appears to be constructed for only one group at a time.
In a number estimation task, subjects judged which of two consecutive displays contained more objects. At the same time, one display had its objects grouped into pairs by either a 1-pixel connecting line, or local similarity of shape, color, orientation, or luminance. Past work showed that when objects were grouped by a line, subjects could not avoid counting groups instead of individual objects, and underestimated the number of objects in the grouped displays. This underestimation helped performance when the smaller display was grouped, and hurt performance when the larger display was grouped.
In the present study, connectivity grouping again led to strong underestimation. Strikingly, none of the similarity grouping cues led to any underestimation. Because number estimation over an entire scene is only affected by grouping processes that operate in parallel, connectivity grouping must be processed in parallel. In contrast, we argue that similarity grouping is serially constructed for one group at a time, via attentional selection of the dimension of similarity. Counterintuitively, this predicts that when we attend to a group of red objects, the grouping among a set of green objects falls apart. This dissociation has been invisible with past tasks based on introspection, which can only reveal the nature of the currently attended group.