A Detection of Deep Fake in Face Images Using Deep Learning

Abstract

Fake images are one of the most widespread phenomena that have a significant influence on our social life, particularly in the world of politics and celeb. Nowadays, generating fake images has become very easy due to the powerful yet simple applications in mobile devices that navigate in the social media world and with the emergence of the Generative Adversarial Network (GAN) that produces images which are indistinguishable to the human eye. Which makes fake images and fake videos easy to perform, difficult to detect, and fast to spread. As a result, image processing and artificial intelligence play an important role in solving such issues. Thus, detecting fake images is a critical problem that must be controlled and to prevent these numerous harmful effects. This research proposed utilizing the most popular algorithm in deep learning is (Convolution Neural Network) to detect the fake images. The first steps includes a preprocessing which start with converting images from RGB to YCbCr color space, after that entering the Gamma correction. finally extract edge detection by entering the Canny filter on them. After that, utilizing two different method of detection by applying (Convolution Neural Network with Principal Component Analysis) and (Convolution Neural Network without Principal Component Analysis) as a classifiers. The results reveal that the use of CNN with PCA in this research results in acceptable accuracy. In contrast, using CNN only gave the highest level of accuracy in detecting manipulated images.