3D Human Face Analysis via Geometrical Descriptors

This study deals with 3D human face geometrical formalization and its applications. Differential Geometry stands as a common background for the whole work. Twelve geometrical descriptors coming from this field and suitable for human face description have been discovered and tested. These descriptors are the six coefficients of the fundamental forms, principal curvatures, Gaussian and mean curvatures, shape and curvedness indexes. By mapping their implementable forms to 3D facial point clouds, a comprehensive formalization of the human face is gained. The first application concerns landmarking. A landmark is a facial point with a geometrical and biological meaning. We will only deal with soft-tissue landmarks, which lie on the skin and are commonly used by maxillofacial surgeons for surgery practices. Landmarks are also largely employed in the field of security, for authentication and identification processes, although at the moment in this field bi-dimensional images are still preponderant. An automatic landmark extraction procedure, here called landmarking, consists in automatically localizing these landmarks on a 3D face. We have worked both on adults' and foetuses' faces and developed two different landmarking procedures. These algorithms have been tested on 3D faces acquired in our department through laser scanner and on 3D ultrasounds obtained through a collaboration with the Hospital of Senigallia. Furthermore, a collaboration with the Department of Computing of Imperial College London has allowed us to develop a Local Binary Patterns- and Histogram-based algorithm, relying on geometrical descriptors, for landmark extraction. This has been tested on the public FRGC database. Other applications concern face recognition and prenatal diagnosis. An algorithm for face recognition has been developed relying on the geometrical descriptors. This is a preliminary method which has been tested on our face dataset. The algorithm of prenatal diagnosis has been designed to detect cleft lip on foetuses and can support ultrasonographers in their work, as it also provides practitioners with a quantification of the defect.