Abstract
Ensemble perception exploits visual redundancy to quickly and accurately extract summary statistics across multiple visual domains. While ensemble processing has typically been investigated using features, objects, and faces, no known experiment has explored the summary statistical coding of sets composed of multiple scenes, despite a neuroanatomical link between scene and ensemble perception (Cant & Xu, 2012). The present study examined summary statistical processing of scene ensembles using two well-established global scene properties: Scene content (i.e., the perceived naturalness or manufacturedness of a scene) and spatial boundary (i.e., how open or closed a scene appears). In a pilot study, participants gave ratings of scene content and spatial boundary for single scenes, and these ratings were used to derive predicted values of average scene content and spatial boundary for scene ensembles presented to a new group of participants in the main experiment (Leib et al., 2016). Within this experiment, we varied set size (n = 1, 2, 4, or 6 scenes in an ensemble) and presentation duration (125, 250, 500, and 1000 ms) and asked participants to rate the average content and spatial boundary of scene ensembles. We found that participants’ ratings were well correlated with the predicted ensemble values at all durations, demonstrating that ensemble statistical processing can occur for multidimensional high-level stimuli, and that such scene-ensemble processing occurs both efficiently and accurately in a fraction of a second. Moreover, we found that participants were integrating more than just 1 scene into their average ratings, as the correlation between actual and predicted ensemble ratings significantly increased across increasing set sizes. These findings demonstrate that statistical summaries can be extracted not only for features, objects, and faces, but also for complex visual scenes. This is consistent with the finding that the processing of ensembles and scenes are mediated by shared neural substrates.
Acknowledgement: Supported by an NSERC Discovery Grant to J.S.C