Xiaojing Zhu¹, Dirk B. Walther¹, Claudia Damiano¹; ¹University of Toronto

Humans can quickly judge simple stimuli as curvy vs. angular, and tend to find curved contours, lines, and shapes more pleasant. This effect extends to architecture photographs, where spaces with curvilinear designs are perceived as more aesthetically pleasing than rectilinear ones, suggesting that curvature information can be extracted from 2D images of 3D spaces. However, image-computable measures of curvature do not correspond to subjective ratings of curvature for complex real-world scenes. We conducted an eye‐tracking study using a 2 (space: curved vs. angular) × 2 (furniture: curved vs. angular) design to determine which visual information people use when making curvature judgments of real-world scenes. Each participant (N=25) viewed 100 indoor images while their eye movements were tracked, and they rated the curvature of scenes on a 5-point Likert scale. We found that participants rated scenes with curvy backgrounds (β = 1.45, p < 0.001) and foregrounds (β = 1.25, p < 0.001) as more curved than angular ones. A wider distribution of fixations, larger saccade amplitudes, and higher saccade velocities were each linked to a small but significant decrease in perceived scene curvature (all p < 0.05). Moreover, curvy backgrounds were associated with lower average saccade amplitude (β = -0.46), saccade velocity (β = -6.62) and fixation spread (β = -13.92), while curvy foregrounds were also linked to smaller amplitude (β = -0.23), velocity (β = -3.15) and fixation spread (β = -6.54; all p < 0.005). In summary, our findings indicate that the presence of curviness in both the background and foreground is linked with higher levels of curvature perception and reduced visual exploration, with the background information playing a more important role in curvature judgments. This is the first step in understanding how humans judge curvature from 3D spaces.