Recognizing degraded faces: the contribution of configural and featural cues.

Human face-recognition processes must maintain high levels of performance under different viewing conditions. An important dimension of variability is image resolution, which is affected by distance, refractive errors, and light levels. Here, we investigate how changes in resolution modulate the visual-system's ability to detect featural versus configural changes in face images. It has been suggested that at lower spatial frequencies the visual system relies predominantly on configural information, yet, to our knowledge, no experiments have systematically examined this idea. We determined subjects' relative sensitivities to configural and featural changes for systematically degraded images. We show that overall configuration and local features are processed equally well at the different resolution levels, supporting the idea of a holistic face-representation that encompasses both feature shape information and information about the distance between the features. These data have also enabled us to derive lower bounds for the resolution needed to effectively use each type of information. Our data are replicated with a completely different face stimulus set, but are not replicated when subjects were shown houses instead of faces. Overall, these results suggest that at lower spatial frequencies, facial representations embody both configural and featural attributes equally, and provide a platform for investigating the essence of holistic facial representations for low-resolution images.