Abstract
Unfamiliar face images often lead to the erroneous categorization of variations as distinct individuals (Jenkins et al., 2011). This tendency is presumed to arise from the multifaceted features inherent in faces, which naturally exhibit variations. Hence, encountering natural variations in a person, such as viewing the person from different angles, under various lighting conditions, with different facial expressions, or makeups, can handle this difficulty through which we identify stable, unchanging features across multiple instances. Different studies on face learning have emphasized the necessity of conceptual learning which involves direct supervision or label information over mere perceptual learning, especially for better generalization on recognizing new instances of learned identities. In the current study, we investigated how exposure to a same person’s variable images via incidental learning, without explicit supervision, influences the perception of newly encountered variations of a learned face—particularly when integrating dissimilar images into a unified identity. In the learning phase, target identities were intermittently presented amid a lot of other distractor faces, with variations within the same identity. Before and after the learning phase, we asked participants to approximate the number of distinct identities among 24 face images from two different people as soon as possible within 8 seconds. Participants either performed the task observing the encountered exemplars during the learning phase or novel exemplars of the learned identities. When comparing the estimated number of identities before and after learning, a prominent decrease in the number was observed after learning. This suggests that the incidental, perceptual learning enables rapid readout of abstract information about specific identities from multiple images. Moreover, the number was smaller in the novel-exemplar condition than the old-exemplar condition. These results indicate that people can equally generalize to new faces through perceptual learning, as long as a variety of instances is presented during learning.