Purchase this article with an account.
Mariia Bulatova, Igor Utochkin; Parallel vs. sample-based extraction of summary statistics from feature and conjunctive sets. Journal of Vision 2014;14(10):1055. doi: 10.1167/14.10.1055.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The visual system efficiently extracts summary statistics (the average or approximate numerosity) from multiple objects at a brief glance instead of encoding individual features (Alvarez & Oliva, 2008; Ariely, 2001). Treisman (2006) claimed that this statistical processing involves globally distributed attention. Following her Feature Integration Theory, Treisman predicts that summary statistics can be derived from feature-marked sets as they are driven by early feature maps, but it is not the case for conjunction-marked sets requiring focused and severely limited attention for feature binding. Treisman confirmed this prediction in a proportion estimation experiment. Previously, we replicated that experiment with subsequent modifications (Bulatova & Utochkin, 2013) and found that observers can be as good at estimating the proportions of conjunctions as at estimating feature proportion, when the number of concurrently presented subsets is controlled properly and not exceeding the limits of working memory capacity. In our current experiment, we tested the mechanisms of this statistical representation. Observers were presented with sets of red, green, or blue T's, X's, and O's and had to evaluate the percent of a precued or postcued feature (either color, or shape) or their conjunction. The spatial distribution of relevant items could be either even over the field, or uneven. We found that spatial distribution only slightly affected the accuracy of feature estimation supporting the notion of massively parallel statistical processing. However, uneven distribution substantially impaired the accuracy of proportion estimation for conjunctive sets. It indicates the failure of parallel processing of such sets. It appears that observers focused their attention on few sample items within a limited region and approximated their estimate to the entire visual field (Myszek & Simons, 2008) which is a good strategy for even but not uneven distributions. Overall, our conclusions are consistent with the Treisman's (2006) account.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only