1. Facial expression editing based on discrete shape space.
- Author
-
LIU Na, WU Zhongke, WANG Xingce, ZHANG Dan, and LÜ Chenlei
- Subjects
FACIAL expression ,GENOME editing ,EXTRAPOLATION ,CARICATURE ,PROBLEM solving ,INTERPOLATION - Abstract
To solve the problems in facial expression editing originated from the influence of the face pose, the small sample data-bases such as sketch or Caricature databases, and the insufficient training data, this paper proposes a Riemann framework based on discrete shape space. In this framework, we can not only resolve the data limitations, but also realize expression interpolation and extrapolation using nonlinear method. The key idea off this paper is building a representation using facial landmarks based on discrete shape space. Built upon the proposed representation, 3D modeling is realized based on an image in the shape space. Then, the 2D caricature or sketch expression extrapolation process can be controlled by the 3D model reconstructed from the input image, and the exaggerated expressions of the caricature or sketch mages generated based on the extrapolated expression of a 3D model that is robust to fad poses in the Kendall shape space. The reason lies in that the Kendall shape space can effectively remove rigid transformations, such as transition, sating and rotation. The experimental results demonstrate that our method can effectively and automatically extrapolate fatal expressions in caricatures or sketches with high consistency and fidelity. Furthermore, compared with the deep earning method, our approach can avoid the construction if complicated training sets for caricature or sketch. [ABSTRACT FROM AUTHOR]
- Published
- 2023