Discrimination of facial identity based on simple contrast patterns generated by shading and shadows

• Published in: Vision research (Oxford) 2023-11, Vol.212, p.108307-108307, Article 108307
• Main Authors: Peterson, Lindsay M., Susilo, Tirta, Clifford, Colin W.G., Palmer, Colin J.
• Format: Article
• Language: English
• Subjects: Face recognition; Illumination; Mooney faces; Shading; Shadows
• ISSN: 0042-6989; 1878-5646
• DOI: 10.1016/j.visres.2023.108307

•Faces can be discriminated using basic visual patterns inherent in facial shading.•Discrimination of these patterns correlates with general facial recognition ability.•Faces illuminated from above are more easily recognised than those lit from below.•Variation in the horizontal lighting direction worsens recognition performance. The pattern of shadows and shading across a face is determined partly by face shape and may therefore provide a cue for facial recognition. In this study, we measured the ability of human observers to discriminate facial identity based simply on the coarse pattern of contrast produced by the interaction between facial geometry and lighting direction. We used highly realistic 3D models of human heads to create images of faces illuminated from different horizontal and vertical directions, which were then converted to two-tone images (‘Mooney faces’) to isolate the coarse pattern of contrast. Participants were presented with pairs of two-tone faces and judged whether it was the same person in both images. Participants could discriminate facial identity based on the minimal cues within the two-tone images, though sensitivity depended on the horizontal and vertical lighting direction. Performance on the Mooney recognition task correlated with general facial recognition ability, though the role of face-specific processing in this relationship was not significant. Our results demonstrate that shading information in the form of simple contrast cues is sufficient for discriminating facial identity, and support the idea that visual processing is somewhat optimised for overhead lighting – here, in the relatively high-level context of face identity recognition.