Objective In this investigation, the precision of a commercial three-dimensional digitizer in the detection of facial landmarks in human adults was assessed. Methods Fifty landmarks were identified and marked on the faces of five men, on five women, and on a stone cast of the face of one man. For each subject, the three-dimensional coordinates of the landmarks were obtained twice using an electromagnetic three-dimensional digitizer, and the duplicate digitizations were superimposed using common orientations and centers of gravity. Metric differences between homologous landmarks were assessed, and Dahlberg's error was computed. Results For both men and women, the error was 1.05% of the nasion-mid-tragion distance, while for the cast, it was 0.9%. When the duplicate digitizations were used to mathematically reconstruct the faces, and several distances, angles, volumes, and surfaces were computed, more than 80% of the measurements had coefficients of variation lower than 1%. Conclusions The digitizer can assess the coordinates of facial landmarks with sufficient precision, and reliable measurements can be obtained.
Guided Spatial Transformers for Facial Expression Recognition
