Comparative Analysis on Different DeepFake Detection Methods and Semi Supervised GAN Architecture for DeepFake Detection

The use of deep learning results in solving a wide range of real-world problems and applications, but there are some drawbacks along with this positive side. One of the most recent and advanced problems among them is the wide use of deepfakes. Deepfakes are digital tampered images or videos created using different deep learning methods. In a deepfake, the face of a targeted person is superimposed on a source image so that this digital tampered data can be used for digital frauds, blackmailing, pornography etc. With the developments in the deep learning field, it is becoming challenging to distinguish between real and fake manually. So it is essential to do research and development in the area of deepfake detection. In this paper, an extensive discussion and timely overview on different deepfake detection methods are done under the classification of feature-based, temporal-based, and deep feature-based deepfake detection. The comparison study is mainly done based on the key features used, face detection architecture, deep learning architecture, video-based or image-based, the dataset used, frames size, and dataset size used. Along with the comparison, a semisupervised GAN architecture is also proposed and developed to detect the deepfake images.