Abstract
Many influential models of scene perception treat objects as the basic unit of scene recognition. However, there is evidence that global properties of an image can drive scene perception before any objects can be identified (e.g., Greene & Oliva, 2009), and computational models suggest this ability could be explained by sensitivity to global patterns across an image (e.g., Oliva & Torralba, 2001). To examine this possibility more directly, we asked whether there is a link between how proficient observers are at representing global patterns (spatial ensemble statistics) and their ability to perform rapid scene recognition. To measure ensemble perception, participants performed a change-detection task in which a grid of Gabors changed orientation, with some changes altering the global pattern of a display (ensemble-different) and others leaving the global pattern the same (ensemble-same; Alvarez & Oliva, 2009). To the extent that participants are sensitive to the ensemble, it should be easier to notice changes on ensemble-different trials than ensemble-same trials (ensemble-benefit). To measure rapid scene recognition, we briefly presented objects with consistent backgrounds vs. inconsistent scene backgrounds, and observers had to report the identity of the object (Davenport and Potter, 2004). To the extent that participants are better at rapid scene recognition, they should have higher accuracy for consistent backgrounds than inconsistent backgrounds (context-benefit). By comparing relative performance scores (ensemble-benefit versus context-benefit), we controlled for factors such as motivation and general object processing skill in accessing ensemble and scene processing abilities. We found a significant correlation between ensemble processing scores and rapid scene perception (N=50; r=0.46; p=0.001): Observers who made the best use of ensemble representations were also most influenced by scene backgrounds in an object recognition task. These results suggest that spatial ensemble representations may underlie rapid scene recognition.
Meeting abstract presented at VSS 2014