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."
