Regions of symmetry in images of natural scenes and artworks

While established as an important contributor to shape or pattern “goodness,” mirror symmetry has yet to be extensively studied in complex images more closely resembling our everyday visual surroundings. With this work, we provide insights into symmetry detection and its underlying processes using a novel bounding-box method on 200 images of natural scenes and 200 images of artworks. In a large online study, 901 participants were asked to indicate regions of symmetry within each image by placing a rectangular bounding box on the image, adjusting it to the region of symmetry they perceived, and marking one of the two axes of the rectangle as the main symmetry axis. They could do this for as many boxes as they saw fit, and they could also use the same bounding box to mark both orthogonal axes in case of double symmetries. For every box, participants also rated the saliency of the region as well as the perceived strength of the symmetry. Next to these behavioral tasks, we asked participants in a separate block to rate the aesthetic appeal of the images. Our new method replicates the previously established dominance of vertical axes of symmetry in dot-patterns and geometric shapes in real-world images. Moreover, we noted a decrease in salience, symmetry strength, and size with the number of boxes within an image. We present the results in relation to our hypotheses and discuss their contribution to the broader literature on symmetry and aesthetics, including a comparison between human and machine vision. Download paper here