Abstract
Our visual system groups areas of the world that share common features. One potential mechanism for similarity grouping is global selection of the similar feature (red, horizontal, square, etc). Grouping a set of red objects might be the same as selectively enhancing areas of an image that contain red, and the same type of process might occur for shape or orientation grouping (e.g. enhancing regions of high curvature or horizontality). The exciting prediction from this account is that only a single group could be constructed at any given moment - the red or the green - but not both simultaneously. This prediction starkly contrasts with the intuition that similarity grouping is parallel. Past work using visual search tasks suggests that similarity grouping is indeed serial. Here we tested further implications of this mechanism: if similarity grouping is created through global feature-based attention, then grouping speed should not be influenced by (1) spacing among elements constructing groups, or (2) how many elements there are in each group (group size). Experiment 1 shows that searching is serial for an unmatched pair of colored squares among matched pairs and vice versa. Moreover, search rates do not differ for groups of widely spaced elements or tightly spaced elements (approximately 50ms/group). Experiment 2 tests whether the number of elements per group affects the time required to construct each group. Consistent with our prediction, searches for particular groups made of four objects were no slower than searches for two-object groups. Collectively, the current results suggest that grouping by similarity is serial, irrespective of spacing or group size. This supports the surprising account that similarity groups are constructed by selecting similar features in a serial manner.
Meeting abstract presented at VSS 2014