Classification of Deepfake Videos Using Pre-trained Convolutional Neural Networks

Abstract

The advancement of Artificial Intelligence (AI) has brought a revolution in the field of information technology. Furthermore, AI has empowered the new applications to run with minimum resources and computational cost. One of such applications is Deepfakes, which produces extensively altered and modified multimedia content. However, such manipulated visual data imposed a severe threat to the security and privacy of people and can cause massive sect, religious, political, and communal stress around the globe. Now, the face-swapped base visual content is difficult to recognizable by humans through their naked eyes due to the advancement of Generative adversarial networks (GANs). Therefore, identifying such forgeries is a challenging task for the research community. In this paper, we have introduced a pipeline for identifying and detecting person faces from input visual samples. In the second step, several deep learning (DL) based approaches are employed to compute the deep features from extracted faces. Lastly, a classifier namely SVM is trained over these features to classify the data as real or manipulated. We have performed the performance comparison of various feature extractors and confirmed from reported results that DenseNet-169 along with SVM classifier outperforms the rest of the methods.