3D Gaussian Splatting in Face Reconstruction

3D face reconstruction is extensively applied across various fields, including virtual reality(VR), augmented reality(AR), computer animation, medical imaging, and face recognition. However, traditional 3D reconstruction methods are hindered by high computational resource consumption, complex processing, and challenges in capturing facial details accurately and efficiently. In recent years, deep learning techniques have significantly improved the accuracy and efficiency of 3D face reconstruction. Nevertheless, challenges remain in real-time rendering and detailed feature recovery. This has been a key focus for current research achieving a balance between efficiency and high accuracy. The main issue in this dissertation relies in the study of 3D Gaussian Splatting (3DGS) techniques for the representation of facial detail and their potential in the optimization of computational efficiency, and the use in digital human models. Experimental results show that 3DGS can enhance the quality of 3D face reconstruction and effectively alleviate the accuracy and computational burden limitations in the traditional methods. This dissertation offers a novel approach and technical pathway for 3D face reconstruction, with significant application value and development potential.