Subitizing And Counting: Preattentive And Attentive Processing In Visual Enumeration

Subitizing, the process of visual enumeration when there are fewer than four items, is rapid (40-100 msec/item), accurate and effortless. In contrast, counting, the process of enumerating more than four items is comparatively slow (250-350 msec/item), effortful and error prone. Why does this occur? In this paper an attempt is made to incorporate subitizing and counting into a general theory of visual perception and spatial attention, as espoused by Marr(1982), Ullman(1984), and Treisman(1988). In particular, it is argued that the rapid apprehension of number in the 1-4 range is parasitic on a preattentive limited capacity mechanism that individuates feature clusters by assigning spatial reference tokens or FINSTs to them(Pylyshyn, 1989). These spatial reference tokens permit the identities of a small number of items to be maintained though their properties and retinal coordinates change, a capability important for directing the attentional focus and coordinating eye and hand movements. If the subitizing process makes use of such preattentive information, then it should not be possible to subitize when spatial attention is required to compute spatial relations, resolve the item as a whole or discern items to be counted from other distractor items. Thus, it was predicted that the slope of the latency function in the 1-4 range should approximate that of the 5+ range if spatial attention is required to perform a particular enumeration task. In contrast, it was predicted that subitizing should be possible when preattentive information could be used to distinguish the items to be counted from one another, or from other distractor items. Therefore, it was predicted that there should be discontinuities in slopes of the latency function between the 1-4 and 5+ range, as shown by deviations from linearity in trend analysis.;Five experiments were performed. In the first, subjects were shown capable of subitizing when the task was to enumerate items of a particular colour though they were not capable of subitizing when the task was to enumerate items that were connected to each other by a contour. This result was expected because spatial attention is presumed necessary to compute the connected relation(Ullman, 1984; Jolicoeur, 1988). The second pair of experiments showed that though subjects can easily subitize when items are defined by groups of contours instead of simple edge points, they cannot easily enumerate such items if they are concentric, as would be predicted given that preattentive grouping processes would cluster all the contours into a unit in this case. The fourth and fifth experiments show that subjects can subitize certain target items in a field of distractors, but only if the property that differentiates targets from distractors is a feature, or a property thought to emerge preattentively. In situations where attention is required to form a unified object description by joining different dimensions (e.g., colour and orientation), or by joining different parts of an item (e.g., an O and a stem to form a Q), subitizing was not apparent. Overall, these experiments suggest that the subitizing process relies on preattentive information