Simulating and validating facial expressions using an anatomically accurate biomechanical model derived from MRI data: Towards fast and realistic generation of animated characters

Abstract

A detailed high-order (cubic-Hermite) finite element model of the human head was constructed from anatomical data segmented from MR images. The model includes a superficial soft-tissue mesh (skin, subcutaneous layer and superficial musculo-aponeurotic system), 20 muscles of facial expressions and several deep structures. Based on the theory of finite deformation elasticity together with large deformation contact constraints, quasi-static facial expressions were generated by activating transversely isotropic muscles that were accurately depicted by their respective anatomical geometries. Material (muscle-fat) heterogeneity was also introduced to capture the realistic formation of skin folds. Using the described approach, four facial expressions were simulated and compared with the surface data obtained from a 3D structured-light scanner. Predicted expressions showed good agreement with the experimental data.