Purchase this article with an account.
Igor Utochkin; How visual set statistics adjust an ‘attentional window’: An information theory of visual search. Journal of Vision 2012;12(9):723. doi: 10.1167/12.9.723.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Experiments of the last decade showed that preattentive system can construct statistical representations of sets (Ariely, 2001) and subsets (Chong & Treisman, 2005a) of objects. To compute statistics visual system should evaluate instances of objects and probabilities of their occurrence in a set. The knowledge of instances and probabilities appears sufficient to compute information entropy of a set, as proposed by Shannon (1948). The theory proposed here suggests that preattentively computed statistics can adjust attentional window in visual search through entropy calculation. This brings new insights into classical visual search phenomena. For example, entropy calculation revealed that attentional capacity of a feature visual display should be just a bit smaller (for sets of under 16 items) or even larger (sets of over 16 items) than a real set size. This implies near parallel mode of attentional processing that fits ‘pop-out’ pattern of feature search. For conjunction displays, entropy calculation revealed rather constant capacity of 1-1.3 items at once irrespectively of set sizes that corresponds to serial search. Another explained phenomenon is attentional capture by a singleton (Theeuwes, 1991). The entropy of a singleton set is larger than the of one without a singleton and this implies narrower attentional window yielding slower target detection. Finally, the theory suggests an account of similarity effects on visual search (Duncan & Humphreys, 1989). If preattentive vision tends to average similar objects under one distribution and separate dissimilar objects into distinct distributions (Chong & Treisman, 2005b) then entropy of a dissimilar set should be larger than of a highly similar set. Hence, attentional window is wider when dissimilar target is to be detected among similar distractors. When similar target is among similar distractors they tend to be averaged under the same distribution. Here, entropy is near zero, preattentive control of attentional window fails and top-down control is predominantly involved.
Meeting abstract presented at VSS 2012
This PDF is available to Subscribers Only