Abstract
Previous work in visual crowding has demonstrated that grouping (e.g. Gestalt laws of grouping) between target and flankers leads to decreased performance (more crowding) than when flankers do not group with the target (Manassi et al, 2012; Chakravarthi & Pelli, 2011). In these experiments, however, the strength of grouping was limited by the requirement that the flanker identities be uninformative for the task, such that knowing the identity of a flanker does not help identify the target. For example, the letter array ANC has less grouping than the array NNN, and knowing the flanker identities in the first case does not explicitly help identify the target. Our recent model of crowding (Balas et al, 2009), which postulates that crowding is a consequence of a high-dimensional pooling representation in the visual periphery, interestingly predicts a reduction in crowding for some situations in which the flankers are informative. Here we test whether grouping between informative flankers and the target is advantageous. To control for informativeness of the flankers (and avoid observers doing the task with the less-crowded flankers) we cross flanker informativeness (ANC vs. NNN) with target-flanker grouping by feature similarity, for several established classes of stimuli (letters, Gabor patches, etc.) and features (orientation, color, etc.). We find that for informative flankers (NNN condition), performance is better when the target shares the same feature as the flankers than when the target and flankers are different along that feature. By treating crowding as an artifact of a compressed peripheral representation, the role of grouping in crowding can be examined in more detail. This research has implications for understanding crowding in natural scene perception, where flankers are more likely to be informative about the target.
Meeting abstract presented at VSS 2014