Abstract

Over the past few decades, artificial intelligence (AI) has been developing rapidly. It has applications in various fields for countless purposes. But not all AI products have a positive impact on society. Sometimes, technologies created for good reasons are later abused by criminals. An example of such technology is known as Deepfake. Deepfake is a technology that allows one to replace a face or change the mouth movement of a person and facial expressions to make them say whatever they wish using any of the deepfake manipulation techniques which include NeuralTextures, Face2Face, Deepfakes, and FaceSwap. Despite its usefulness, if used maliciously, it can severely impact society, for instance, spreading fake news and cyberbullying, among others. Using Deepfakes manipulation techniques, this study, therefore, aim to address this problem by proposing a model that analyses the frames of a video using a deep learning approach to detect the forged areas in the video and to deploy the trained model using Flask (a framework used for the deployment of web applications in python). The proposed model uses an EfficientNet B6 classifier to train a neural model to detect deepfake images via a FaceForensics++ dataset. The trained model was able to classify videos with an accuracy 90%. To validate the model performance, the precision, recall and f1-score was utilized with a result of 98%, 81% and 89% respectively for the class of fake images and a result of 84%, 98% and 91% for the class of real images. The motive of the study was to improve learning by this model.