Study of Segmenting Hair Regions by Deep Learning for Automatic Portrait Generation

Abstract

In this paper, in order to perform automatic portrait generation, we consider a method of automatically identifying and extracting hair area from face image using deep learning. By learning a Convolutional Neural Network (CNN) using teacher data whose label image s of hair region s created manually are output s and front al facial RGB images are inputs , Semantic segmentation is performed to predict categories other than hair area or hair area. As a result, we obtained a prospect of automatically extracting the area of hair with high precision even for unknown face images.