Video Deepfake Detection

Abstract

"Multiple approaches have been followed for Deepfake detection, the most promising being FST-matching, Siamese Neural Network and Recurrent Neural Network (RNN). FST-matching involves comparing source(original) to the Fake (potential manipulated content) in videos to detect alterations. It solely depends on inconsistencies between two of such facial landmarks making it a viable option for Deepfake detection. Siamese Neural Networks involve in identifying dissimilar inputs on a pair of images or videos that are mapped father apart in manipulated content. RNN is a Neural network that will be used to identify temporal information present in videos to certain patterns that are prone in Deepfake. All these methods are viable options for Deepfake detection, and the system aims to make use of both to increase robustness and accuracy."