Abstract
Extending ideas abstracted from the visual systems' ability to efficiently summarize and compress information, we investigated how pairing estimates from individual observers based on their patterns of performance across the visual field can optimize performance in real world detection tasks. Observers completed several detection tasks in which they were presented with very brief (500 ms) images of 9 objects and asked to indicate the extent to which they thought each image should be "called back" for further inspection based on its likelihood of containing a pre-specified target. Strategically pairing observers based on their patterns of performance in detection tasks involving targets defined by spatial frequency, contrast, orientation, and/or size, then averaging estimates from the resultant pairs on an otherwise identical task where they had to detect the presence of a tool among other objects resulted in marked improvements in detection performance (d') over estimates taken from individual observers (all ps < .001), especially in a rare target condition where the tool target was only present on 5% of trials. Results suggest promising new methods of efficiently combining independent estimates to maximize detection performance within limited pools of observers, such as experts who must monitor images for important targets like weapons in baggage scans.
Meeting abstract presented at VSS 2017