Abstract
Local symmetry, measured via the medial axis, has been shown to influence behavior on visual tasks, from the detection of objects in noise, the classification of objects, up to the classification of scenes. Electrophysiological and fMRI studies have suggested various visual brain areas that might encode local medial symmetry. Information about the medial axis impacts the neural representation of objects as early as V3. Neurons in V4 are selective for specific locally symmetric shape parts. The local symmetry of a scene is encoded in the visual representations of PPA. Here, we wish to shed further light on where in the visual pathway local symmetry is represented for natural stimuli. We used the Bold5000 dataset, which contains 5000 images along with BOLD responses to each of those images for 4 participants. We algorithmically scored the local symmetry content of each image from the dataset. We regressed the mean local symmetry score of each image, the mean luminance, and the contrast against the BOLD activity. Whereas luminance and contrast affect activity in early to mid-level visual areas (V1-V4), local symmetry influenced neural activity in mid to high-level areas (V4, LOC, PPA, and RSC). As most of the previous work investigating symmetry in fMRI focused on global symmetry on dot patterns, we wanted to look at the effect of global symmetry on real-world scenes. To this end, participants on MTurk rated the amount of global symmetry in each image. We regressed their ratings to the BOLD responses from BOLD5000. Similar to local symmetry scores, the participant ratings influenced activity in LOC, suggesting that this area might represent both global and local symmetry. Unlike local symmetry, global symmetry did not significantly influence PPA activity. Our data shows, for the first time, that local and global symmetry differentially influence neural responses to real-world scenes.