Detection of Deep Fake in Face Images Based Machine Learning

Abstract

Fake face images is a recent critical issue of artificial intelligent due to it has directly impacts on the social lives, and may be made to imply threats against privacy, fraud, and other issues. Currently, creating fake images has become relatively simple due to the powerful yet user-friendly mobile applications that navigate in the social media world and with the invention of the Generative Adversarial Network (GAN) that provides a good quality  images that might be difficult for humans to differentiate with their eyes and makes image and video manipulation simple to do, quickly spread and hard to detect, Therefor, image processing and artificial intelligence are crucial in solving such problems. That is why scientists must create technologies or algorithms to control and to avoid these various negative impacts by different detection approaches can be applied. The proposed approach is more robust than current methods when propose a model based on support vector machine as a classifiers to detect fake human faces created by machines. The first stages includes a preprocessing that start with changing images from RGB to YCbCr and then applying the gamma correction. finalize the results show that the extracted edges using Canny filter were useful for detecting fakes in face images. After that, applying two distinct methods of detection by utilizing "Support Vector Machine" with "Principal Component Analysis" and "Support Vector Machine" without "Principal Component Analysis" as a classifiers. The findings show that the highest accuracy gained is 96.8% when using the SVM with PCA while the accuracy obtained is 72.2% when using the SVM alone.