The Deepfake Challenges and Deepfake Video Detection

Abstract

Deepfake is a combination of fake and deep-learning technology. Deep learning is the function of artificial intelligence that can be used to create and detect deepfakes. Deepfakes are created using generative adversarial networks, in which two machine learning models exit. One model trains on a dataset and then creates the deepfakes, and the other model tries to detect the deepfakes. The forger creates deepfakes until the other model can't detect the deepfakes. Deepfakes creating fake videos, images, news, and terrorist events. When deepfake videos, and images increase on social media people will ignore to trust the truth. Deepfakes are increasingly affecting individuals, communities, organizations, security, religions, and democracy. This paper aims to investigate deepfake challenges, and to detect deepfake videos by using eye blinking. Deepfake detections are methods to detect real or deepfake images and videos on social media. Deepfake detection techniques are needed original and fake images or video datasets to train the detection models. In this study, first discussed deepfake technology and its challenges, then identified available video datasets. Following, convolutional neural networks to classify the eye states and long short term memory for sequence learning has been used. Furthermore, the eye aspect ratio was used to calculate the height and width of open and closed eyes and to detect the blinking intervals. The model trained on UADFV dataset to detect fake and real video by using eye blinking and detects 18.4 eye blinks per minute on the real videos and 4.28 eye blinks per minute on fake videos. The overall detection accuracy on real and fake videos was 93.23% and 98.30% respectively. In the future research and development needs more scalable, accurate, reliable and cross-platform deepfake detection techniques.