Identifying DeepFake Videos Using Multimedia Processing Techniques

Abstract

Abstract: Different deepfake software can be used to create fake videos which seek spreading fake news and misinformation. It is considered as one of the problems that needs to be combated nowadays.Even though creating a high-quality fake video to fool people is both expensive and time consuming but it is not difficult, and many researchers succeed in constructing realistic fake videos. This resulted in large number of fake videos on the internet, hard to distinguish, and highly misleading to ordinary public.Therefore, in this research we proposed a new technique to detect Deepfake videos accurately by applying different multimedia processing techniques to find lips movements and compare it to the associated speech signal.Two modules are used which are Voice Activity Detection (VAD) and Mouth Openness and Closeness Detection (MOCD) Modules. VAD module is analysing the audio features and plot the speech signal; while, MOCD module is analysing the video frames to find the mouth region and calculate its ratio aspect. The outputs of both modules are compared together to check if the silent intervals are matching the mouth closeness or not. If the speech signal is mismatching the mouth movements; then, the video is fake. Testing the modules are carried out using a collected dataset of real and fake videos and it’s obtained 85.7% accuracy.Using the proposed technique in detecting fake videos will decrease its sinister side. Also, it will improve the current security systems by preventing criminals from building fake videos to use it for authentication purposes.