This study reports on the design and evaluation of a tool to assist in the description and visualisation of the human face and variations in facial shape and proportions characteristic of different ethnicities. A comprehensive set of local shape features (sulci, folds, prominences, slopes, fossae, etc.) which constitute a visually-discernible ‘vocabulary’ for facial description. Each such feature has one or more continuous-valued attributes, some of which are dimensional and correspond directly to conventional anthropometric distance measurements between facial landmarks, while other attributes capture the shape or topography of that given feature. These attributes, distributed over six facial regions (eyes, nose, etc.), control a morphable model of facial shape that can approximate individual faces as well as the averaged faces of various ethnotypes. Clues to ethnic origin are often more effectively conveyed by shape attributes than through differences in anthropometric measurements due to large individual differences in facial dimensions within each ethnicity. Individual faces of representative ethnicities (European, East Asian, etc.) can then be modelled to establish the range of variation of the attributes (each represented by a corresponding three-dimensional ‘basis shape’). These attributes are designed to be quasi-orthogonal, in that the model can assume attribute values in arbitrary combination with minimal undesired interaction. They thus can serve as the basis of a set of dimensions or degrees of freedom.

The space of variation in facial shape defines an ethnicity face space (EFS), suitable for the human appreciation of facial variation across ethnicities, in contrast to a conventional identity face space (IFS) intended for automated detection of individual faces out of a sample set of faces from a single, homogeneous population. The dimensions comprising an IFS are based on holistic measurements and are usually not interpretable in terms of local facial dimensions or shape (i.e., they are not ‘semantic’). In contrast, for an EFS to facilitate our understanding of ethnic variation across faces (as opposed to ethnicity recognition) the underlying dimensions should correspond to visibly-discernible attributes. A shift from quantitative landmark-based anthropometric comparisons to local shape comparisons is demonstrated. Ethnic variation can be visually appreciated by observing the changes in a model through animation. These changes can be tracked at different levels of complexity: across the whole face, by selected facial region, by isolated feature, and by isolated attribute of a given feature. This study demonstrates that an intuitive feature set, derived by artistically-informed visual observation, can provide a workable descriptive basis. While neither mathematically-complete nor strictly orthogonal, the feature space permits close surface fits between the morphable model and face scan data. This study is intended for the human visual appreciation of facial shape, the characteristics of differing ethnicities, and the quantification of those differences. It presumes a basic understanding of the standard practices in digital facial animation.