A computational method for bilateral symmetry recognition in asymmetrically scanned human faces

One of the most important geometric features of the human face is the plane identifying its bilateral symmetry. Thanks to the possibility of having 3D data of the face, the bilateral symmetry plane can be recognised and evaluated by means of properly processed face data. The knowledge of the symmetry plane could be useful for various purposes, such as: - face authentication ([1] and [2]); - face reconstruction or cosmetic surgery for aesthetic corrections in Maxillofacial Surgery ([3] and [4]); - facial asymmetry and symmetry line correlation for back pathologies in Orthopaedics and Orthodontics [5]; - facial symmetry and cognitive disorders correlations for schizophrenia diagnosis in Neurology [6]. In many practical cases the data relating to an acquired human face are incomplete and not symmetrically acquired. These limitations are mainly due to the typical technologies used in 3D geometric scanning. These technologies require that each and every part of the face should be completely visible from the device viewpoint. Considering the anatomy of the face, many parts may result in undercut and therefore may not be acquired. The portion of the face which can actually be scanned depends on the face’s distance from and orientation with respect to the scanning device. In real cases bilateral symmetry must be recognised on a partially acquired surface, which is also affected by a non-uniform point density. Thus the information which attempts to recognise human face symmetry must be extracted from an asymmetric point cloud. Other factors which may blur the symmetry recognition process can be identified in some physiognomic asymmetries of the human face, such as local damage, pimples, bumps, wrinkles, etc. Typically, the methods presented in literature are conceived and tested to analyse cases where the face is completely acquired. They fall short in the case of asymmetrically scanned data and whenever there is a non–uniform sampling density. With a view to detecting the symmetry plane of asymmetrically scanned human faces, this paper presents a new method. Its performance, in terms of robustness and accuracy, is quantified as to the symmetry plane detection of some real test cases.