Computational modeling of face recognition based on psychophysical experiments

Recent results from psychophysical studies are discussed which clearly show that face processing is not only holistic. Humans do encode face parts (component information) in addition to information about the spatial interrelationship of facial features (global configural information). Based on these findings we propose a computational architecture of face recognition, which implements a component and configural route for encoding and recognizing faces. Modeling results showed a striking similarity between human psychophysical data and the computational model. In addition, we could show that our framework is able to achieve good recognition performance even under large view rotations. Thus, our study is an example of how an interdisciplinary approach can provide a deeper understanding of cognitive processes and lead to further insights in human psychophysics as well as computer vision.