Abstract
Humans are incredibly fast and accurate in detecting faces across different scene contexts. While scene information has been shown to affect object detection, it is unclear whether it also modulates the rapid detection of faces. To find out, we presented participants (n=18) with 200 natural target scenes and recorded their eye movements during a face detection task. Each target scene came from a different natural indoor setting (e.g., bathroom, basement) and contained a single face. For each scene, we created a corresponding face-less version by manually editing the image. Prior to each target scene, participants either saw an initial preview of the face-less version or a gray screen (no preview) for 250ms. Which half of the images was shown with preview was counterbalanced across participants. We found that participants’ first saccade on the face was earlier in the face-less scene preview compared to the no-preview condition (mean time to fixation: 235 vs. 250ms, p=0.02). Participants were also faster in indicating they found the face via button-press in the face-less preview condition (mean RT: 1018 vs. 1088ms, p=0.00). These findings show that prior information about a scene facilitates human face detection behavior, placing important constraints on bottom-up accounts of face perception.