Fighting Deepfakes Using Body Language Analysis

Recent improvements in deepfake creation made deepfake videos more realistic. Open-source software has also made deepfake creation more accessible, which reduces the barrier to entry for deepfake creation. This could pose a threat to the public privacy. It is a potential danger if the deepfake creation techniques are used by people with an ulterior motive to produce deepfake videos of world leaders to disrupt the order of the countries and the world. Research into automated detection for deepfaked media is therefore essential for public safety. We propose in this work the use of upper body language analysis for deepfake detection. Specifically, a many-to-one LSTM network was designed and trained as a classification model is trained for deepfake detection. Different models trained using various hyper-parameters to build a final model with benchmark accuracy. We achieve 94.39% accuracy on a test deepfake set. The experimental results show that upper body language can effectively provide identification and deepfake detection.