Makeup Transfer Using Support Vector Regression
Tsuji Ayumi, Masataka Seo, Yuko Muto, Yen‐Wei Chen · 2018
This study aimed to generate a virtual makeup facial image considering personal facial features (texture information) using machine learning. In conventional makeup simulator systems, the makeup image (target image) is simply transferred to the user's facial image, and they do not consider the user's individual facial features. Therefore, in this study, we used image pairs of unpainted faces and their makeup faces developed by beauticians as training data to learn a mapping function between the natural image and the makeup image using support vector regression (the mapping function represents the experience of beauticians). Subsequently, using the estimated mapping function, we automatically generated a virtual makeup image based on the individual features of an unpainted (natural) facial image (the user's input image). Additionally, we extend our method to transfer the makeup texture to a three dimensional reconstructed facial image so that the user can objectively evaluate the differences in the impression of makeup from different viewpoints.