Chapter 6 - A review of techniques to detect the GAN-generated fake images

Abstract

Recent advancements in the domain of artificial intelligence and computer vision has given the computing devices the capabilities to mimic the human traits. These days, artificial intelligence has given capabilities to machines to book flight tickets by just talking over the phone, do medical diagnosis, Siri and Alexa can act like our virtual assistants and many things can be done by machines by just giving a set of instructions. Although the innovations carried out in this domain have been quite beneficial, these advancements have also created ways and means by which mischiefs can also be done. The technology has advanced to such an extent that we can even create fake images that are almost similar to the real ones, the alteration is done in such a way that the fake images go unnoticed by the human eye, for example, forgery. One of the most famous and reliable approach based on artificial intelligence to generate fake images is generative adversarial networks (GANs). It has the capability to change and manipulate the pixels in a novel way to create images that never existed. These techniques have popularized over the years and are generally used to create fake images. The GAN techniques are backed by two neural networks: one of them is the generator network and the other one is the discriminator network. The generator network is responsible for creating the images and the discriminator network takes care of the training images, thus it tries to find out which images are the real images and which images have been created using the generator network. Both the generator and the discriminator networks play a minimax game, in which one tries to increase the probability of being correct whereas the other tries to minimize the probability of being right. In recent days, the fake images thus created have flooded the internet with misinformation and have also created privacy concerns. These images are generally uploaded over the social media to promote the fake news and to increase its trustworthiness. Many frauds can also be carried out by using this unlawful and unethical means. Therefore, many researchers have contributed in recent years to develop methods that can detect the authenticity of the images and can easily distinguish between the authentic and the fake images, thus created with the help of technology. The contributions made in this context will be quite helpful to curb this menace of falsehood and hence enable us to protect the privacy of the individual. In this chapter, an effort is being made to compare the various state-of-the-art techniques that have been proposed so far to detect the fake images generated by the GAN technology and to find out the research gaps that can be further improved upon.