Data-Driven Geometric Face Image Smilization Featuring Moving Least Square Based Deformation

Sijie Shen, Toshihiko Yamasaki, Kiyoharu Aizawa, Toru Sugahara

Abstract: With the prosperous development of the Internet, virtual identity is becoming more important to people in social life. As one key factor, attractiveness, is a significant part of virtual identity, where there is still room for different strategies to improve attractiveness. In this study, we focus on digital face image and make it a little bit smiling as a common way to improve attractiveness. The workflow includes face detection, face alignment and image deformation. By only deforming geometrically, our method reaches the smile rate at 53.96% while natural smile image has 89.29% smile rate, based on crowd sourcing experimental results on our test set.