Facial Attractiveness Ontology Using 3D Craniofacial Morphable Sythesis

Langlois et al. argued that averaged faces were perceived as attractive in their pioneering findings. The debates of “beauty” or “attractiveness” and its relationship with races, genders, or even health, have since staged sporadically. Recent studies further promoted the importance of facial image averaging in face recognition, plastic surgery, and dentistry. The performance, in terms of recognition accuracy and visual satisfaction, of image averaging is, however, greatly dependent of the employed algorithmic approaches. In this 2-year project, we propose an eigenface-based approach by using the hybrid of traditional craniofacial landmarks and facial action units to achieve morpphable averaging wherein preserving consistent features of the face. We develop the concept of the three-dimensional weighted-averaging lattice (WAL) which constructs the attractiveness hierarchy with the higher-level synthesized faces being the average of their connecting lower-level faces – the more towards the hierarchical root, which is the ultimate average of all, the more attractive the face is, based on the argument by Langlois et al. In this project the lattice will serves as the instrument to help guide planning and prediction of the craniofacial surgery. The physician can follow along a surgical planning path on the lattice to design its operative strategy, and to predict the pre- and post-surgery changes. To evaluate the efficacy of a craniofacial surgery, we propose to rank the post-surgical facial image in terms of the number of levels “uplifted” in the WAL: an image-based closest matching is performed between the post-surgical face and the images at all levels in WAL. Further, Balanced Angular Profile Analysis (BAPA) is proposed to perform the anthropometry. O The project will be carried out in two years. The aims of the research and the anticipated results would include: (1) Propose the weighted-averaging lattice (or WAL) and weighted-averaging faces (or WAF) based on local facial features (2) Study of Facial Attractiveness Ontology: Collaborate the Analytic Hierarchy Process (AHP) and Fuzzy Neural Networks to correlate facial features and facial attractiveness, with application of questionnaire. (3) Establish Three-Dimensional Morphable Synthesis Model using the weighted-averaging lattice. To facilitate the collection of 2D/3D facial images, we shall acquire images with the help of the 3DMD and Cone-beam CT mechanism of Chang Gung Memorial Hospital. In case of any difficulty, we shall collect facial images from publicly available Face Recognition (http://www.face-rec.org/databases), GavabDB, and USF Human ID 3D databases

Project ID:PB10107-1722
External Project ID:NSC101-2221-E182-013