Visual search is a daily task for all of us, from finding our car keys to looking for a colleague in a crowd. Given the importance of this task, much research has been devoted to it, and thus we know a great deal about visual search in artificial two dimensional displays. However, visual search in the real world occurs in complex, yet highly structured three dimensional environments. What are the principles that guide search in real-world scenes? A separate line of research has highlighted role of contextual regularities between objects and scenes. In other words, knowing that a keyboard is found in offices helps the recognition of both keyboards and offices. Do such regularities help guide attention in real-world visual search problems? While the importance of these statistical regularities has been widely acknowledged, they have not been measured or quantified. It is necessary to measure these regularities to understand the role that they play in search. Here, we have amassed a large scene database of 3500 scenes and have completely measured all objects and regions in these scenes. This rich dataset includes information on what objects occur in different scene categories, and the spatial distributions of the objects'positions. We propose to analyze this dataset to extract statistical regularities existing between objects and their scene context, as well as regularities from the co-occurrence structure between objects. We will use the formal framework of information theory to quantify the degree of regularity in these relationships. This allows us to put an upper bound on the amount of guidance we can expect from these statistics. Then, we will perform behavioral experiments examining the use of these statistics in real-world visual search problems. These data allow us to ask questions and make predictions that have previously been impossible, therefore allowing real-world search to be studied in natural scenes in a controlled and principled way. Health relevance Understanding how attention is deployed in real-world visual search tasks has many public health implications. Understanding difficult visual search problems could lead to better accuracy at interpreting x- ray and MRI data, as well as search for abnormalities during endoscopic surgery. Furthermore, understanding search can help aid those whose search abilities are compromised due to visual (e.g. macular degeneration) or attentional (ADHD or age related cognitive decline) reasons.

Agency
National Institute of Health (NIH)
Institute
National Eye Institute (NEI)
Type
Postdoctoral Individual National Research Service Award (F32)
Project #
5F32EY019815-04
Application #
8142799
Study Section
Special Emphasis Panel (ZRG1-F12A-E (20))
Program Officer
Steinmetz, Michael A
Project Start
2009-10-01
Project End
2012-09-29
Budget Start
2011-09-30
Budget End
2012-09-29
Support Year
4
Fiscal Year
2011
Total Cost
$51,326
Indirect Cost
Name
Stanford University
Department
Type
DUNS #
009214214
City
Stanford
State
CA
Country
United States
Zip Code
94305
Greene, Michelle R (2016) Estimations of object frequency are frequently overestimated. Cognition 149:6-10
Greene, Michelle R; Baldassano, Christopher; Esteva, Andre et al. (2016) Visual scenes are categorized by function. J Exp Psychol Gen 145:82-94
Iordan, Marius C?t?lin; Greene, Michelle R; Beck, Diane M et al. (2016) Typicality sharpens category representations in object-selective cortex. Neuroimage 134:170-179
Iordan, Marius C?t?lin; Greene, Michelle R; Beck, Diane M et al. (2015) Basic level category structure emerges gradually across human ventral visual cortex. J Cogn Neurosci 27:1427-46
Greene, Michelle R; Liu, Tommy; Wolfe, Jeremy M (2012) Reconsidering Yarbus: a failure to predict observers' task from eye movement patterns. Vision Res 62:1-8
Greene, Michelle R; Wolfe, Jeremy M (2011) Global image properties do not guide visual search. J Vis 11:
Wolfe, Jeremy M; Vo, Melissa L-H; Evans, Karla K et al. (2011) Visual search in scenes involves selective and nonselective pathways. Trends Cogn Sci 15:77-84