A Comparative Analysis of Deep Learning Based Approaches for DeepFake Identification

One of the major concerns in the present global environment is the creation and utilization of deepfakes. In this work, we have examined the issues and challenges that are created by the deepfake in the security and surveillance mechanisms. Deepfakes are used for by passing the facial identification or biometrics system that are normally used as surveillance mechanisms. Previously the only issue was to find the authentic person through the photographs that they have put on their identity proofs but nowadays it is extended to find the forged pics that are manipulated with the usage of AI based methods like deepfakes. Deepfakes are used in the current environments for bypassing the security by impersonation and providing false information and thus become a source of danger especially in the politics arena and entertainment landscape. Deep learning has the tendency to mitigate the impact of deepfakes to considerable extent by incorporating the same in the techniques develop for the identification of deepfakes. This paper provides the comprehensive review of the deepfake related work that has been done by researchers with the usage of deep learning approaches which could identify the fake images, videos to significant extent in various contexts along with the mechanism used for deepfake creation and identification in general. Moreover, the comparative analysis of the existing techniques has been done considering a range of factors like dataset used, technique used, accuracy, AUC, data type used for deepfake identification etc. It has been inferred that majority of the researchers have used the FaceForensics++, Celeb-DF, DFDC dataset and have utilized CNN technique for the deepfake identification primarily on images.