A Survey for 2D and 3D Face Alignment
Xiaofei Hui · 2019
Face alignment, which is critical for many face applications such as face recognition, is to detect facial landmarks automatically. Existing methods can be classified into two categories: 2D solution and 3D solution. There are many advanced methods in 2D solution to optimize the detection on the image plane. 3D solution employs a 3D dense face model for 3D facial landmarks estimation. Though the two solutions have made great progress, there are limitations hindering face alignment accuracy's further improvement. 2D solution lacks the ability to process faces in large poses like profile face, while 3D solution is not superior in alignment of common-angle images like near-front faces. In this paper, a comparative survey is presented to provide a holistic understanding of both 2D and 3D face alignment, which may inspire more advanced solutions.