Abstract
Faces are extremely important stimuli for humans, and there is growing evidence that face detection and localisation occurs early in the visual hierarchy. Faces provoke saccadic responses within 100 ms after stimulus onset (Crouzet et al., 2010, J Vis), and face-selective neural representations have been identified in early visual areas (Campana et al., 2020, bioRxiv). Given that face detection may be accomplished by early, low-level visual processing, we examined whether detection and localisation can be accomplished in the absence of perceptual awareness. In a 2AFC saccadic choice task, faces and objects were presented for 8, 100, or 400 ms, followed by 400 ms phase-scrambled masking images. Observers were asked to look for the face (or object) target and to make a manual response corresponding to the target location. We predicted more accurate saccadic responses to faces compared to objects, independent of stimulus visibility. Results showed high detection accuracy even at the 8 ms SOA, with only a 20-30% reduction in visibility for face and object targets, respectively. Eye-tracking data showed that saccadic response was not necessary for accurate detection. In the critical 8 ms SOA condition, observers only made saccades on about 20% of the trials despite high manual detection accuracy. However, observers were more accurate in detecting faces, and when they did execute saccades, saccadic accuracy towards targets was above chance for faces but not for objects. These results demonstrate a remarkable robustness for face detection with very little visual input and without initiating saccadic response. Better understanding of the covert detection processes for such limited visual input may extend models of human eye movements for overt visual targeting.