Detection of artificial images and changes in real images using convolutional neural networks

Abstract

Image recognition has now become one of the most popular methods used in the entertainment industry, media, automotive, etc. The possibilities provided by neural networks and deep learning algorithms, cause the development of various methods for generating, modifying and falsifying information. An example would be the use of deep learning algorithms to replace faces in a video recording. Social networks, video materials are full of fake video and images. Our work proposes a method of detecting forgery on real images and detecting artificially generated images using Convolutional Neural Networks (CNN). Our approach introduces the possibility of classifying images into one of three classes: the class of real images, the class of real and modified images, and the class of artificially generated images. An important element of our work is the practical detection of modified or artificially generated images that could be used when phishing biometrically protected data. The research has been narrowed down to the facial images of people of different skin colour, nationality and age. The conducted tests show the acceptable effectiveness of our method and become a positive element of further experiments.