Abstract
To recognize an object we first detect features and then combine or “pool” them. Crowding, the inability to identify a peripheral object surrounded by flankers, is thought to be a breakdown of feature pooling. Thus, it has been widely used to study object recognition. Features within objects are usually spatially distributed. How much do the flanker features furthest from the target affect crowding? Do features need to be part of an unbroken object to be combined with target features? We create a set of six target objects consisting of jagged vertical sides attached by flat tops and bottoms. These objects differ only in the features on their jagged sides. All flankers are derived from this set. We use whole objects as flankers in one condition, and broken objects as flankers in others. We break an object by introducing a gap between its jagged sides, violating object closure. At various eccentricities, we present a target object between two flankers. We vary the width of the flankers by keeping the near jagged side a constant distance from the target while moving the far jagged side (for unbroken objects, top and bottom stretches, for broken objects, size of gap increases). We find that, in the whole object condition, as the flanker's far side moves closer to the target, crowding increases. However, for broken objects, moving the far side closer to the target does not affect crowding. Further, there was little crowding with broken objects. We ruled out target-flanker similarity as an explanation. Our finding that the far side of a flanker combines with the target object only when it is attached to the near side suggests that objects, not features, pool. We conclude that whether a flanking feature is combined with the target depends on what flanking object the feature belongs to.