Abstract
Ensemble perception is the visual system’s ability to take advantage of the redundancy of the world by representing complex environments via summary statistics. Ensemble representations are used not only for low-level features like orientation and size, but for high level features, like face identity and emotion (Haberman & Whitney, 2009). While previous research has found that we are still able to derive the ensemble when faces are partially artificially occluded via amodal bars (Haberman & Ulrich, 2019), it is unknown how ensemble representations are impacted when faces are occluded by more naturalistic objects, such as face masks. To test this we had observers complete a continuous report task with faces that were partially occluded with standard surgical masks. Participants (N=19) were shown four faces simultaneously for one second and then were asked to select the average identity of the four faces on a continuous report wheel. In one block, these were faces without masks, and in another faces with masks. We also had a single face condition with both unmasked and masked blocks. The absolute error in degrees around the face wheel from the true average face was our measure of how accurately subjects could derive the ensemble face or represent the single face. We found that participants were equally able to extract the ensemble of a masked group of faces as they were for unmasked faces. This result demonstrates that ensemble representations for the identity of faces can still be derived with naturalistic occlusions. Overall, this suggests that our visual system may look beyond occlusions and accurately calculate summary statistics of faces that are not visible in their entirety.