Abstract
Research of the past decade has shown that many features of multiple objects can be efficiently represented as summary statistics, e.g. average size. Statistical properties are supposed being hierarchical features affecting global similarity and the perceived properties of individual objects (Ariely, 2001; Brady & Alvarez, 2011). They presumably can affect, therefore, the property known as salience, an ability of visual singleton to capture attention in bottom-up fashion. We tested this hypothesis in 4 visual search experiments. We proposed that the salience should increase with the distance from ensemble average defining global similarity. We also proposed that salience is affected by local similarity defined by the number of items closest to a target by critical dimension. Observers searched for a size target among 13 or 25 items of four different sizes. 3 target sizes were intermediate between adjacent distractor sizes, 2 other targets were outside distractors size distribution, close to smallest and largest distractors. A target between 2nd and 3rd distractor sizes was an average, a most globally similar target. In Experiments 1 and 2 observers searched for an odd size target either in sets with normally (Exp. 1), or bimodally (Exp. 2) distributed sizes. The distribution shapes allowed testing local similarity effects as a function of number of adjacent distractors. We found in the result that within distractors distribution search efficiency didn’t depended on global or local similarity; it depended solely on absolute size demonstrating typical size search asymmetry. Outside targets were, however, much better detectable. In Experiments 3 and 4, observers searched for a known size targets, and results were substantially the same. Hence, both global and local similarities appear to have no effect on target salience within items distribution, perhaps, due to noisy interactions insufficient for representing individuals. However, boundary targets gain greater salience, perhaps, because of being treated as outliers.
Meeting abstract presented at VSS 2013