Generation And Detection of Deepfakes using Generative Adversarial Networks (GANs) and Affine Transformation
J. Vijaya, Amaan A. Kazi, Kishan G. Mishra, Avala Praveen · 2023
Deepfake technology has gained significant attention and notoriety in recent years for its ability to manipulate visual and audio content, leading to widespread concerns about its potential for abuse. Deepfake videos can be created by using deep learning algorithms that enable the synthesis of facial and vocal features of a target individual onto another person, leading to convincing and often misleading videos. While the technology behind deep fakes is constantly evolving, there has been increasing interest in understanding and mitigating their potential negative impacts. Researchers and experts are exploring ways to detect and prevent the creation of malicious deep fakes while also developing ethical guidelines to regulate their use. This paper presents a deep fake project that utilizes Generative Adversarial Networks (GANs) and affine transformations to generate deep fake videos. The proposed approach takes a target image and a driving video as inputs and generates a realistic-looking deep fake video that mimics the facial expressions and movements of the driving video and combines it with the target image and also incorporates a classification model to detect whether the generated deep fake video is real or fake. The classification model will also be based on the same architecture as used in generating the fake videos and will be used to evaluate the realism and quality of the generated deep fake videos. It generates a novel approach for generating and detecting deep fake videos that can be applied to various applications, such as entertainment, education, and security. However, creating deep fakes also offers potential benefits, such as improving the entertainment industry and enhancing the quality of content creation. As technology advances, it is crucial to balance its potential benefits with its potential risks and take appropriate measures to ensure that deep fakes are used ethically and responsibly. The proposed method aims to identify the real identity of a person appearing in a video and use this information to detect whether the video is genuine or not. This identity-aware approach leverages a deep neural network architecture that combines facial recognition and face forgery detection techniques. The proposed solution has the potential to enhance the security of digital media and protect individuals from various forms of identity theft and cyber-crime.