The self as an image: Appearance and belief in visual representations of one's own face

Self-representations are central to cognition, yet their internal structure and visual fidelity remain poorly understood. We combined behavioral and computational approaches to characterize self-face representations from perception and memory by linking similarity judgments, model alignment, and image reconstruction. Participants provided similarity judgments for self, familiar, and unfamiliar faces, which were compared against the representational structure of artificial neural networks (ANNs). We then used behavior-based image reconstruction to visualize the image-level consequences of those similarity structures. Self-face judgments aligned reliably with recognition-trained, identity-separating ANNs, but not with the tested generative latent spaces. Pixelwise similarity also explained unique variance, indicating that these judgments contained both identity-level and pictorial information. Consistent with this pattern, reconstructions recovered self-face information from both perception and memory. These reconstructions preserved identity-relevant structure but were less constrained by the target image than reconstructions of other faces, with image-specific fidelity weakest for recalled self-faces. Exploratory bias analyses suggested that self-beliefs modulate these representations, especially in memory: higher self-rated attractiveness was associated with more attractive reconstructed self-faces, and higher self-concept clarity was modestly associated with more accurate representations. Together, these findings indicate that self-face representations constitute visually recoverable yet systematically biased representations, shaped by both objective appearance and self-belief-related information.