Detecting Deepfakes using CNN and LSTM

Deepfake, a face-swapping method that has been abused recently, has caused a great deal of public worry. Effective countermeasures are required because several deepfake videos, sometimes known as "deepfakes," have already been created and posted online. Therefore, Deepfake detection is going to be a promising defense against deep fakes. It is quite easy to imagine cases in which these convincing tampered videos are exploited to foment political unrest, extort money, or stage terrorist attacks. In this study, a pipeline is suggested for identifying and classifying deepfake videos. Our system uses convolutional neural networks (CNNs) to retrieve frame-level characteristics. Then, a recurrent neural network (RNN-LSTM) is built using these features to determine if a video has been altered or not. We study, analyze and compare various methodologies and ours on a vast cluster of deep fake videos from various video sources. We present how our system can complete this task effectively with comparative results while utilizing a straight forward architecture.