Abstract
A relatively large literature in environmental psychology has been dedicated to the restorative effects of natural scenes, in which the term “restoration” refers to the improved cognitive functioning found in various tasks. The literature on the restorative effects of nature suggests a set of vaguely defined terms that are associated with restoration. Are there any quantifiable measures that these terms correspond to? The purpose of the current study is to examine the relationships between the low-level visual features and terms associated with restoration. Low-level features include statistics like edge density (ED), pixel entropy, saturation, and standard deviation of saturation. The definitions of the terms associated with restoration were adapted from the Perceived Restorativeness Scale. Through Amazon Mechanical Turk, 88 participants rated 680 outdoor images on naturalness and four restorative terms. We found 20 weak but significant correlations. The largest correlations were between naturalness and ED (r=.30), being-away and saturation (r=.25) and being-away and ED (.22). We also trained linear classifiers on low-level features to learn whether or not an image was rated high on each term. Classification accuracy was 72% for fascination, 83% for coherence, 79% for scope, 60% for being-away, and 61% for naturalness. The small correlations that we found suggest that low-level statistics can only weakly predict these terms. However, the above-chance success of the classifiers imply that a non-linear combination of these statistics might be predictive of the terms. We discuss these results in terms of the biases of the databases and question how representative the images are of our experiences with natural scenes.