Stereo-assisted Landmark Detection for the Analysis of 3-d Facial Shape Changes

Techniques for the three-dimensional analysis of facial morphological changes resulting from surgical treatment have recently been proposed. These approaches largely rely on expert manual intervention for placing anatomical landmarks on 3-D facial models which is time consuming and error prone. A new automated approach is presented here which combines a stereo-assisted active shape model for 3-D landmark extraction with geometric morphometrical tools for a statistical analysis of facial shape changes. A stereophotogrammetric imaging system is used for acquiring 3-D face models. This incorporates an active shape detection phase which is used for the automatic localisation of 2-D facial features in greyscale stereo images. The detected features can then be used to generate 3-D soft tissue landmarks by stereo correlation matching and disparity map interpolation. Generalised Procrustes analysis, principal component analysis and thin plate spline decomposition are then applied to the analysis of shape changes in 2-D facial midline profiles and extracted 3-D facial landmarks. The proposed method is validated both statistically and visually by characterizing shape changes induced by mandibular repositioning using a bite block in a heterogeneous sample of 20 patients attending a weekly orthodontic clinic. It is shown that the method is capable of distinguishing between changes in facial morphology due to mandibular repositioning and changes due to other factors such as growth and normal variation within the patient cross-sectional sample.