Attentional selection by distractor suppression.

Selective attention was studied in displays containing singletons popping out for their odd form or color. The target was defined as the form-singleton, the distractor as the color-singleton. The task was to discriminate the length of a longer line inside the target. Target-distractor similarity was controlled using a threshold measurement as dependent variable in experiments in which distractor presence vs absence, bottom-up vs top-down selection (through knowledge of target features), and target-distractor distance were manipulated. The results in the bottom-up condition showed that length threshold was elevated when a distractor was present and that this elevation progressively increased as the number of distractors was increased from one to two. This set-size effect was not accounted by the hypothesis that selective attention intervenes only at the stage of decision before response. Selective attention produced a suppressive surround in which discriminability of neighboring objects was strongly reduced, and a larger surround in which discriminability was reduced by an approximately constant amount. Different results were found in the top-down condition in which target discriminability was unaffected by distractor presence and no effect of target-distractor distance was found. On the other hand, response times in both bottom-up and top-down conditions were slower the shorter the target-distractor distance was. On the basis of the experimental results, selective attention is a parallel process of spatial filtering at an intermediate processing level operating after objects have been segmented. This filtering stage explores high level interactions between objects taking control on combinatorial explosion by operating over only a limited spatial extent: it picks out a selected object and inhibits the neighboring objects; then, non-selected objects are suppressed across the overall image. When no feature-based selection is available in the current behavior, this filtering influences perception in decreasing discriminability of non-selected objects. When feature-based selection is available, spatial interactions are set before stimulus arrival, hence only the unmatching objects have their discriminability diminished.