Deep learning 3D facial reconstruction framework for prosthetic rehabilitation.

Aesthetics has been one of the concepts most studied and analyzed by the greatest philosophers, authors and thinkers of all times and is nowadays a key factor in case of prosthetic rehabilitation, especially when connected to facial rehabilitation (e.g., in case of severe accidents or pathologies). The implementation of the aesthetic standards in daily clinical practice to improve the aesthetics of patients has been a constant challenge for clinicians since dentistry was born. The traditional approach in this domain is to create wax models or similar artifacts that can help the patient to visualize the result if the prosthetic rehabilitation process. The advance in computing graphics techniques, machine learning and artificial intelligence has the potential to significantly impact this domain. The goal of this thesis is to build a complete framework for 3D facial reconstruction in prosthetic rehabilitation by applying cutting edge computing image processing and deep learning techniques. The key functionalities of the framework will be as follows: - capability of reconstructing facial 3D models starting from 2D pictures taken with commercial low cost cameras - capability of clustering 3D facial models of healthy individuals based on a set of facial aesthetics metrics in order to create a library on which training advanced classification models - capability to match the 3D facial model of a patient to a cluster of compatible models in the constructed library - reconstruction of the damaged portion of the patient face by exploiting the matched models. The ultimate goal is to provide the clinicians an instrument to display to a patient the effect of prosthetic rehabilitation.