Component and configural information in view-based face recognition by human and machine

In this study we used the inter-extra-ortho paradigm from Bülthoff & Edelman (1992) in order to investigate what kinds of information are used for recognizing faces across viewpoint. In Experiment 1, ten face-stimuli were presented in frontal view and 45° side view. At test they had to be recognized among ten distractor faces from 13 different viewpoints. We found systematic effects of viewpoint (recognition performance: inter > extra > ortho). In Experiment 2 and 3 we investigated the relative importance of component and configural information for view-based recognition. In Experiment 2, participants learnt 10 intact faces. At test, configural information was destroyed by scrambling faces (10 target faces and 10 distractor faces). The results suggest that faces can be recognized on the basis of isolated featural information. In Experiment 3, 10 previously learnt intact faces were shown among 10 distractors in low-pass filtered versions. This eliminated local features contained in the facial parts resulting in a stimulus containing only configural information. The blurred versions could be recognized reasonably well, which suggests an important role of configural information. Furthermore, we found systematic effects of viewpoint for both isolated configural information and isolated component information. In addition to the psychophysical experiments, systematic differences between the effects of viewpoint are discussed in a computational framework based on key frames