Deep Neural Networks for Multimedia Forensics: Detection of Deepfake Content

Abstract: "Deepfakes" are digital manipulation techniques that use deep learning to create false photos and movies. The most difficult component of obtaining the original is detecting deepfake photographs. Because deep fakes are becoming more widely known, it is critical to distinguish original photographs and videos in order to identify manipulated films. This project investigates and tests several methods for detecting between genuine and fake photographs and movies. Deep fakes were recognized using the Convolutional Neural Network (CNN) approach known as Inception Net. In this research, a comparative comparison of many convolutional networks was performed. This project takes use of a Kaggle dataset containing 3745 images generated during the augmentation process, as well as 401 videos of train samples. The accuracy and confusion matrix metrics were utilized to evaluate the outcomes. The proposed model exceeds the others in terms of accuracy, detecting deepfake photographs and videos with a 93% rate.