DeepFake Detection Using Error Level Analysis and Deep Learning

Abstract

The image recognition software is used in numerous distinctive industries that include entertainment and media. The deep learning (DL) algorithms have been of great help in the development of several techniques used for creating, altering, and locating any data. The deepfake method is a photo-faking technique that includes replacing two people's faces to an extent that it becomes very difficult to identify it with a naked eye. The convolution neural network (CNN) models including Alex Net and Shuffle Net are used to recognize genuine and counterfeit face images in this article. The technique analyzes the performance and working of all distinctive algorithms using the real/fake face recognition collection from Yonsei University's Computational Intelligence Photography Lab. The first step in the process starts by the normalizing of pictures then the Error Level Analysis is carried out before it is put into several difference CNN models. Then the in-depth features are extracted from the CNN models utilizing the Support Vector Machine and the K-nearest neighbor methods. The most perfect accuracy of 88.2% of Shuffle Net via KNN was analyzed while Alex Net's vector had the accuracy of 86.8%.