On Machine Learning and Deep Learning based Deepfake Generation and Detection

With the advancement of artificial intelligence, deepfakes have evolved into a potent tool that allows the developer to manipulate images or audios that can lead to defamation or any other kind of security threat. It is a cutting-edge technology that uses deep learning and machine learning techniques which gives the user enormous power to create deep fake media for both entertainment and malicious purposes that may result in high impact in real life scenarios. Hence, recently the research communities have been increasingly interested in the development of approaches for detecting deepfakes as the trust on the media available online comes under dilemma. In this paper, a comprehensive overview of deepfake technology with its pros and cons, followed by deepfake generation methods like Encoder-Decoder and GAN is discussed. The benchmark datasets with the open-source tools for deepfake generation have also been discussed in detail. How the face manipulation techniques like Face-Swap, Face-Synthesis, Face- Attribute-Manipulation and Face-Re-enactment are used is also a part of this study. Additionally, it offers a comparison of past research on the identification of deepfake images and videos which are applying deep-learning and machine-learning algorithms. The research gaps of this technology and how this can be implemented for further research, perspective, and insights of the same have also been given. An evaluation on the machine-learning and deep-learning based detection models for fake images, videos, audios, and multimodal content has also been explored and presented in this paper.