Video Morphing Attack Detection Using Convolutional Neural Networks On Deep Fake Detection Algorithm

Abstract

A method called deepfake produces fake video and films with artificial or substituted faces. Deepfakes are turning into a worrying societal phenomenon because they may be used maliciously to spread harmful information, fabricate electronic convincing proof, make fake political news, and even participate in online harassment and fraud. Regarding the fight against the ubiquitous threat provided by deepfake videos, our suggested technique, named "Video Morphing Attack Detection Using CNN," is a stronghold. We provide our system the capacity to model temporal relationships across sequences and extract complex characteristics from video frames by integrating the ResNeXt-50 and LSTM neural network designs, respectively. Our method effectively detects abnormalities suggestive of video morphing assaults via forward propagation and SoftMax activation. We guarantee strong detection performance while maintaining the veracity and integrity of visual information by utilizing dynamic face localization and specific assessment measures. Our system provides a comprehensive solution to reduce the negative impacts of deep fake manipulation in visual media, including vital functionalities for picture conversion, prediction creation, and data pretreatment. Convolutional neural network architecture ResNeXt-50 is a member of the ResNeXt family, which was first presented as an advancement of the ResNet design. To overcome certain constraints of conventional convolutional neural networks (CNNs), ResNeXt aims to achieve state-of-the-art performance in image classification tasks while managing intricate visual input. The "50" in ResNeXt-50 stands for the network's depth or the total number of layers it has. ResNeXt-50 comprises 50 layers in total, comprising fully connected, pooling, and convolutional layers.