A Novel Approach for Deepfake Video Detection

As technology continues to progress, the creation of deepfakes is becoming increasingly simple and accessible. Due to developments in deep learning, artificial intelligence, and face manipulation tools, the quality of average deepfakes is steadily broadening. Deepfake technology represents the forefront of video manipulation, employing sophisticated neural networks such as auto-encoders and Generative Adversarial Networks (GANs) to produce convincingly deceptive facial videos. Replacing the target face with a fake face in video content can be achieved quite easily using face manipulation techniques. Deepfakes can deploy the different type of deep learning algorithm's which can easily manipulate and synthesize the images and videos of a person that humans cannot distinguish from the real one. Caution should be exercised when utilizing this, as it has the potential for misuse. Identifying and detecting the manipulated video content is an essential responsibility that requires resilient methods to detect deepfake videos. There are serious concerns that deepfake technology will become a problem in the future. biometric security may become more vulnerable to malicious use in the coming years, raising concerns about its safety. Information, involving the utilization of facial recognition technology. To address these issues, detecting deepfakes are extremely crucial. Therefore, we attempt to construct a hybrid deepfake detection model which can categorize the manipulated videos and original videos. In this paper we will develop more efficient model to detect whether a video is a real or manipulated video. We will be using Res-Next Conventional Neural Network to extract frame level features and Long Short-Term Memory (LSTM) to differentiate whether a video is fake or real type. There are various available datasets to train and test our model. Our main focus is identifying and detecting deepfake videos with high precision, effectively by using the robust method which can differentiate b/w real and deepfake video with accuracy of more than 94%.