AI Based Deepfake Detection

Abstract

Advances in machine learning, especially following the 2014 release of Generative Adversarial Networks, have allowed techniques and methods to be used for nefarious ends. Generative Adversarial Networks can even create fake images and videos which appears to be real to human eyes. Generative Adversarial Networks can swap the faces of two different people. For film producers or graphic designers, this tool is quite useful. Face swapping has been done in movies to replace the real person's face with that of the actor. A computer-generated versions of actors Carrie Fisher and Peter Cushing have been featured in a movie named “Star Wars: The Rise of Skywalker” from the “Star wars” film series just like they appeared in the 1977 original, while other Marvel films have “de-aged” actors such as Michael Douglas and Robert Downey Jr. However, this has the potential to be misused. By producing ultra-realistic Deep Fake videos using various trailblazing machine learning techniques, felon are trying to harass, blackmail the innocents. It can also be used to induce political instability by disseminating erroneous information, which can cause communal, diplomatic, and violent outbreak with disastrous consequences. This gives rise to a significant menaces to security of the person as well as national defence, necessitating the development of automated methods for detecting deep fake videos. The eye blinking pattern in deepfaked videos is not formed as naturally as it should be due to the incapability of Generative Adversarial Networks. This will come in handy when constructing a deepfake detecting algorithm. The project uses an object's eye blinking pattern to determine whether or not a video is deepfaked.