Deepfake Detection in Videos and Picture: Analysis of Deep Learning Models and Dataset

Abstract

Deepfake detection is the concept of distinguishing a computer manipulated graphic from a real recorded graphic. The technology used for this purpose is deep learning. It is a sub branch of artificial intelligence. With technology becoming more readily available, deepfakes are also increasing in use in recent years. It becomes evident that a system is needed that detects deepfakes and prevents its use in suspicious activities. Development of a deepfake detection technology becomes evident to avoid the use of deepfakes in such activities. For this purpose, many tech giants have assimilated huge datasets which consist of videos that were made using deepfakes already available. To detect a deepfake, one requires an equally capable or even better algorithm and detection technique. Generative Adversarial Nets, GANs, is one such technique that might be able to rival other deepfake techniques. This paper will discuss various methods to apply to detect deep fakes along with the process, libraries used, dataset liabilities and limitations, analysis and efficiency. Since Deep Learning technology is evolving each day with new innovations, this paper provides a comparative study about methods that have already been tested and their limitations with respective models and how to possibly make them more efficient.