Fighting Deepfakes Using Body Language Analysis

Recent improvements in deepfake creation have made videos more realistic. Moreover, open-source software has accessible, which reduces the barrier to entry for creation. This could pose a threat people’s privacy. There is potential danger if techniques are used by people with an ulterior motive produce of world leaders disrupt order countries and world. Therefore, research into automatic detection deepfaked media essential public security. In this work, we propose method using upper body language analysis. Specifically, many-to-one LSTM network was designed trained as classification model detection. Different models were varying hyperparameters build final benchmark accuracy. We achieved 94.39% accuracy on test set. The experimental results showed that can effectively detect deepfakes.