Detailed Features-Preserving 3D Facial Expression Transfer

In the 3D facial expression transfer field, aiming at two hot problems of preserving rich detailed information target model to make generated new expressions realistic and natural, reducing training time, this paper presents a features-preserving method. Firstly, features are extracted from face models obtain basic without details. Then, source is transferred with improved parametric dimensionality reduction by unsupervised regression. Finally, restored using proposed feature vector adjustment strategy. The visual contrast quantitative analysis experiments reconstruction accuracy time as evaluation indexes conducted on datasets such COMA in Matlab software under Windows 10 environment. results illustrate that compared nonlinear co-learning method, method can not only losses, but also well preserve personalized model, so it more natural. effectively improves speed transfer.