Automatic Face and Hijab Segmentation Using Convolutional Network

Abstract

Taking pictures and Selfies are now very common and frequent between people. People are also interested in enhancing pictures using different image processing techniques and sharing them on social media. Accurate image segmentation plays an important role in portrait editing, face beautification, human identification, hairstyle identification, airport Surveillance system and many other computer vision problems. One specific functionality of interest is automatic face and veil segmentation as this allows processing each separately. Manual segmentation can be difficult and annoying especially on smartphones small screen. In this paper, the proposed model uses fully convolutional network (FCN) to make semantic segmentation into skin, veil and background. The proposed model achieved an outperforming result on the dataset which consists of 250 images with global accuracy 92% and mean accuracy 92.69.