Abstract
The visual system extracts summary statistics from crowds of similar items. This heuristic, known as ensemble perception, works across multiple visual domains ranging from low-level features such as size, shape, and orientation to high-level objects such as faces, biological motion, and animacy. In the current study, we examine how observers represent the emotional valence of a complex group of scenes. In a pilot experiment observers evaluated the emotional valence of several hundred scenes on a scale ranging from ±5. From these, a total of 190 scenes were selected that spanned the full range of emotional ratings (and had an equal number of positively and negatively rated images). Scenes included pictures of people grocery shopping, working, cooking, etc. Observers were first asked to rate the emotional valence of each scene, and then subsequently rate the average emotional valence of randomly assembled groups of four scenes. Results indicate that there is significant compression of the ensemble ratings relative to the expected rating based on the individual images (i.e., ensembles, regardless of their overall valence, were viewed as less extreme than the individual ratings would predict). These results are surprising for two reasons: 1) they demonstrate ensemble representations for abstract scene information and 2) they contrast with other work suggesting that ensemble ratings for certain objects are amplified relative to individual ratings (e.g., Harp, Haberman, & Whitney, VSS poster, 2009).
Meeting abstract presented at VSS 2017