Zigui Zhu¹, Xiaoxue Gao², Philippe Schyns³, jiayu Zhan¹; ¹Peking University, ²East China Normal University, ³University of Glasgow

Face identification relies on both observers' perceptual abilities and their conceptual understanding of faces. However, it's unclear how these components contribute to the identity information observers memorize to recognize familiar faces. To address this question, we applied a data-driven approach to model, at individual level, the facial features that 48 participants (29 Females and 19 males) used to identify their own faces, and for comparison the facial features they used in face discrimination and facial attractiveness rating. Specifically, we used a generative model of 3D faces to synthesized a broad range of female and male faces that varied naturally in 3D face shapes and complexions. Then, we tested these generated faces in three within-subject tasks: 1) assessing same-sex faces for similarity to their own face in memory, 2) evaluating the physical similarity of paired faces based on perception, and 3) rating opposite-sex faces from "very unattractive" to "very attractive" according to personal preference. We reverse correlated the specific features that modulated memory-based self-face similarity, perceptual-based visual similarity, and concept-based attractiveness. Through pairwise and triple-wise comparisons, we analyzed the specific fit of 3D shape and complexion features across three feature sets, and identified the memory bias linked to individuals’ perceptual sensitivity and conceptual preferences. These biases were further characterized in relation to participants’ face memory abilities and their attitudes toward self-attractiveness. Our findings reveal how memory biases are shaped by perceptual and conceptual processes, offering a novel framework for understanding the idiosyncratic biases in high-dimensional facial feature processing.