Background and Question. A series of studies (Rossion and Tarr, forthcoming; Tarr et al, 2001) demonstrate the diagnosticity of global color information for gender categorization and our ability to exploit this source of information in certain conditions. Our current work looks into the advantage provided by color variation over facial features compared to global color information. The relative utility of surface cues is assessed with respect to the one of geometrical cues. Structural information is deployed for recognition purposes also as a way to establish the utility and psychological plausibility of segmentation-based face representations.

Methods. Pattern classification / recognition, behavioral experiments

Results and Discussion. Hue variation over facial features was found to be a significant predictor of objective gender and of human performance in a gender categorization task. Independently, it was shown with the aid of a reverse correlation method that humans use patterns of color distribution to discriminate between genders.

Current and future work. We examine the diagnosticity of color distribution for face identification and for other categorization tasks. The development of a segmentation-based method for face recognition using color distribution along with other cues is under way. In parallel we attempt to apply reverse correlation to the study of face identification as a way to determine the nature of featural and configural information humans use in face processing.

People. Adrian, Giulia