Proceedings of the Annual Meeting of the Cognitive Science Society

Valentine’s face-space suggests that faces are represented in apsychological multidimensional space according to their per-ceived properties. However, the proposed framework was ini-tially designed as an account of invariant facial features only,and explanations for dynamic features representation were ne-glected. In this paper we propose, develop and evaluate a com-putational model for a twofold structure of the face-space, ableto unify both identity and expression representations in a singleimplemented model. To capture both invariant and dynamicfacial features we introduce the face-space duality hypothesisand subsequently validate it through a mathematical presen-tation using a general approach to dimensionality reduction.Two experiments with real facial images show that the pro-posed face-space: (1) supports both identity and expressionrecognition, and (2) has a twofold structure anticipated by ourformal argument.