Study Of 3D Face Super-Resolution Based On Progressive-Resolution-Chain

As the rapid development of computational and network transporting capability, people are raising their requirement to the quality of multimedia data. The technique of super-resolution aims at improving the quality of the multimedia data such as images and videos. It could increase details or remove blur effect by extracting and fusing the information of the multi-frame inputs or by compensating learned priori knowledge to the single input data.However, the existing super-resolution methods focus on the data of 2D domain, while the algorithms for 3D models are little addressed. Since 3D models have become an important data form of multimedia, 3D super-resolution problem is of great significance for both theory and practical application.In this thesis we start with the survey and comparison among the existing 2D super-resolution approaches. Then, we give the definition of the 3D super-resolution problem, and propose a learning based super-resolution framework for 3D models. The experiment on USF HumanID 3D face database demonstrates the feasibility of the proposed framework and the effectiveness of our algorithm. The main contributions of this thesis are as follows:1. We establish a learning based super-resolution framework for 3D models.2. We propose the Progressive-Resolution-Chain(PRC) model, which connects the 3D models of different resolution using their intrinsic relationship.3. A two-stage mapping using the intrinsic parameterization is established, by which the 3D super-resolution problem is transformed into the one of 2D domain.4. Based on the PRC model, we propose the PCA based 3D Super-Resolution algorithm, and give the whole solution to the problem.