Abstract
Observers can extract the overall “gist” of a scene in the form of summary statistical information about visual features or objects that have explicit dimensionality (e.g., average orientation, size, facial identity or emotion). However, our rich perception of scenes seems to encompass more than this, suggesting that individuals can extract ensemble information from perceptual impressions that are not immediately specified by the visual features in the image. Here we tested whether observers perceive lifelikeness—how “alive” a groups of static, random objects is. In a pre-test, we asked 20 observers to rate the perceived animacy of 150 individual static objects. There was high agreement between observers, and these pre-test ratings were used to generate crowds of random objects containing varying degrees of animacy. In Experiment 1, 20 new observers rated the average animacy of groups of up to 6 objects displayed simultaneously for 1 second. We regressed each observer’s ratings against the predicted crowd ratings generated by the independent observers in the pre-test, and obtained highly significant correlations. These results indicate that observers perceived average animacy, and that overall animacy is predictable from the individual objects comprising the crowd. We replicated these results in Experiment 2, in which participants rated the average animacy of a group of random objects, sequentially displayed for 50 ms each, showing that observers can extract ensemble information about animacy within a fraction of a second. Sub-sampling control conditions confirmed that participants integrated multiple objects into their ensemble percept. Our results demonstrate that ensemble perception is not restricted to features or dimensions that are explicitly available in the image. Instead, ensemble percepts can operate over abstract visual interpretations, providing a link between summary statistical representations of basic visual features and the rich perceptual experience that observers report.
Meeting abstract presented at VSS 2015