Frequency-based Deep-Fake Video Detection using Deep Learning Methods

Abstract

Deep Learning (DL) is an advanced and effective technology widely used in diverse industries, including medical imaging (MI). Data Mining (DM), Image Processing (IP), and Machine Vision (DM). Deep-fake uses DL technology to alter videos to render them indistinguishable from the original humans. The effectiveness of deep-fake has recently obtained significant attention from researchers, and numerous DL-based techniques have been developed to identify deep-fake videos. In this paper, a novel deep-fake video detection method is proposed. The Deep Fake Detection Challenge (DFDC) and Face Forensic datasets were used in the research. In addition, frequency-based frame extraction was conducted on each video during the preprocessing stage. Convolutional Neural Networks (CNN) Long Short-Term Memory (LSTM) - CNN  techniques were used to identify fake videos. The LSTM-CNN approach achieved an accuracy of 82%. To identify fake videos using DL techniques, this work will be helpful to researchers.