Events > Human face perception
Prof. Dr. Michael A. Webster; University Nevada, Reno
How the visual system encodes information about faces remains very poorly understood. In this talk I will consider some of the general coding principles that have emerged from studies of color vision, and whether these can provide insights into the mechanisms of face perception. These include principles about the ways in which spectral information is organized (e.g. norm-based coding and spectral opponency) and about what information is represented (e.g. color constancy and color categories). Studies of color vision have also been important for understanding how visual coding might be optimized through both long-term and short-term adaptations to the observer's environment, and I will explore the implications of analogous adaptations in face perception.
Prof Dr. Volker Blanz; Universität Siegen
Morphable Models of 3D Faces are a representation of natural shapes and reflectances (textures) of human faces learned from a dataset of 3D scans. In this representation, each face is a vector in a high dimensional face space, and new faces can be generated by linear combinations of examples. In computer vision, Morphable Models can be used for face recognition, for model-based reconstruction of 3D faces from single images, and for rotating faces in images. These automated systems have demonstrated the viability and the significant benefit of learning class-specific information from training examples and relying on this prior information in subsequent vision tasks. After a brief summary of recent work in computer vision, we turn to another relevant application of Morphable Models: By generating controlled stimuli for experiments, Morphable Models provide a unique tool to explore biological vision. Faces can be morphed continuously, attributes such as gender or attractiveness can be learned from examples and subsequently modified in given faces, and faces can be turned into caricatures or morphed beyond the average to generate anti-faces. We present some recent psychological findings in perceptual adaptation to anti-faces across changes in head pose which suggest that the locus of adaptation is on a high-level, viewpoint-independent stage of neural processing.
Dr. Isabelle Bülthoff; Max-Planck-Institut für biologische Kybernetik
Humans can easily identify faces at the individual level although faces belong to a class of objects with high similarity between exemplars. Characterizing conditions for which faces are more easily recognized allows us to better understand the mechanisms underlying face recognition.
Numerous studies have shown that distinctive faces are better recognized than typical faces. Those results have implication for the mental representation of faces. In a set of experiments we tested cross-modal effects of distinctiveness. More specifically we asked whether distinctive voices can improve memory for otherwise typical faces. Our results suggest that the quality of information in one modality, i.e., audition, can affect recognition in another modality, i.e., vision; thus showing that face distinctiveness can be of multi-modal nature. Because we encounter faces of only two sexes but recognize faces of innumerable different identities, it is often implicitly assumed that sex classification is an easier task than identification. We investigated how sensitive we are to variations of identity-related features or sex-related features of highly familiar faces. The results suggest that while extracting and processing sex-related information from a face is a comparatively easy task, we do not seem to retain sex-related facial information in memory as accurately as identity-related information. These results have implications for models of face representation and face processing.