Effects of Glow Data Augmentation on Face Recognition System based on Deep Learning

Abstract

Biometric artificial intelligence application depends on amount of material on which they are trained. In this paper, we integrated Glow data augmentation technique to diversify the facial images dataset to analyze its effects on face classification and identification system based on Convolutional Neural Network (CNN). In first phase, we trained our CNN with publicly available Labeled Faces in the Wild (LFW) database and evaluated the proposed system, which achieved accuracy of 92.2%. In second phase, we diversified LFW dataset with Glow method and then trained our CNN network. The experiment results shows that Glow data augmentation improved the accuracy of proposed network to 93.6%.