Data Augmentation Techniques for Facial Image Generation: A Brief Literature Review

Abstract

Image processing has gained notoriety over the last few years in performing various tasks through deep learning (DL) algorithms, such as face recognition and identity verification. Unfortunately, most of them require a large set of images for training, usually manually labeled, which is a costly task both in time and effort, not to mention being prone to human error. Data Augmentation (DA) techniques have been used to mitigate this situation, as they generate images by applying variations to real image sets. This chapter presents a brief literature review on variousDAmethods dedicated to image generation. The technique that has presented outstanding results in the task of generating artificial images is Generative Adversarial Networks (GANs). Some recent research papers in which GANs have been used for the generation of artificial images are presented. General aspects of GANs, such as their definition, architecture, training, and challenges, are described. Additionally, the implementation of aGANarchitecture for the generation of artificial face images from a public set of images is presented. The need for a great computational capacity to generate images with great sharpness and realism is highlighted.