A Review on Various Deepfakes’ Detection Methods

In this era of fake digital content, deep generative models have lately demonstrated outstanding outcomes in a variety of real-world applications like open access to large-scale public databases. Moreover, we can generate high-resolution and diverse samples with the help of advanced deep learning techniques, particularly Generative Adversarial Networks (GANs). This results in the development of remarkably realistic fake content, causing significant concern and spreading distrust in multimedia content with related societal impact. Hence it arises the urgent need for automated methods to detect fake multimedia generated by artificial intelligence. Although many face editing algorithms appear to produce realistic human faces, closer inspection reveals aberrations in specific domains that are typically invisible to the naked eye. These deep learning-based contents are knowns as deepfakes. There are four broad categories of deepfakes that are as follows: photo deepfakes, audio deepfakes, video deepfakes, and audio–video deepfakes. This paper provides a review of the existing generation and detection methods of the various deepfake contents. It also provides a detailed comparison of the objectives, methodology, and algorithms proposed in various studies by different researchers in recent years. Finally, the paper concludes with the notion that one should minimize the restrictions and bottlenecks that are experienced by the existing methods by proposing more advanced techniques of detection.