Semantic, Statistical and Aesthetic Determinants of How Natural and Urban Images Make Us Feel

Introduction: Natural environments are consistently more preferred and perceived as more restorative than urban environments. Studies have shown that certain high-level semantic characteristics of natural environments (presence of water and green vegetation), low-level image properties (color, edge density, spectral slope, entropy) and perceived aesthetics are all associated with increased ratings of preference and perceived restorativeness. However, these factors have been studied in isolation of each other with their relative contribution in driving the perception of natural and urban images still unknown. Method: We investigate how the perceived restorative properties of natural (N=200) and urban (N=152) images are influenced by a variety of semantic subcategories (Water, Greenery, Street, Building or Architecture) and the associated low-level physical properties including chromatic (HSV) and spatial characteristics (amplitude spectrum slope, fractal dimension, entropy, edge density, and directionality). The images were rated for perceived complexity, beauty, liking and visual interest. To assess perceived restorativeness we asked participants to report how engaged, calm, and refreshed the images make them feel. In experiment 1, a total 638 participants (Amazon MTurk) viewed and rated natural (N=318) and urban scenes (N=320) separately on one of the seven rating scales with at least 42 participants per rating scale. In experiment 2, additional 362 participants (UNSW SONA Participant Pool) rated all natural and urban scenes on one of the rating scales with at least 40 participants per rating scale. Results and Conclusions: We found large and significant differences in the low-level physical properties not only between natural and urban scene categories but between the semantic subcategories of each scene type. While image category (natural vs urban) and the respective semantic sub-categories were strong predictors of perceived aesthetic and restorative properties, partial correlation and regression analyses, showed that low-level image properties were independent predictors of both perceived aesthetic and restorative properties.