Abstract :
[en] Face learning studies suggest that exposure to image variations—such as changes in lighting, viewpoint, background or physical appearance—can facilitate recognition. While variations in lighting and viewing angle are known to support image-independent representations, the roles of variations in background or physical appearance remain unclear. Real-world evidence indicates that stable appearance may aid early learning, with variability becoming helpful later. To fill that gap, we conducted two pre-registered experiments examining how appearance (stable vs. variable hair colour and style) and context (stable vs. variable background scene) influence face learning. In Experiment 1 (N = 84), participants learned eight new faces, each introduced through separate blocks of six images—a standard lab-based method. Recognition was tested after a 20-min delay, using images that displayed only inner facial features. We hypothesized that stability would encourage a coarse encoding while variability would help focus attention on diagnostic inner facial features, improving performance. Results showed that variation in appearance, but not background, enhanced learning of inner features. Therefore, variability in physical appearance had more influence on learning than background changes. In Experiment 2 (N = 84), the same images were presented in a random order, mimicking real-world exposure. Here, we expected stability—in appearance and background—to support identity matching during learning and encoding of diagnostic facial features. However, neither factor significantly impacted recognition. Comparing the two experiments, performance for stable and variable faces in Experiment 2 fell between the highest performance for variable faces and the lowest performance for stable faces in Experiment 1. This suggests that random presentation may diminish the benefit of appearance variability while improving performance for stable faces. These findings imply that the blocked format commonly used in face learning studies may inflate the benefits of variability, potentially diverging from learning processes observed in more naturalistic settings.
