Deep learning has been used in computer vision to accomplish many tasks that were previously considered too complex or resource-intensive to be feasible. One remarkable application is the creation of deepfakes. Deepfake images change or manipulate a person’s face to give a different expression or identity by using generative models. Deepfakes applied to videos can change the facial expressions in a manner to associate a different speech with a person than the one originally given. Deepfake videos pose a serious threat to legal, political, and social systems as they can destroy the integrity of a person. Research solutions are being designed for the detection of such deepfake content to preserve privacy and combat fake news. This study details the existing deepfake video creation techniques and provides an overview of the deepfake datasets that are publicly available. More importantly, we provide an overview of the deepfake detection methods, along with a discussion on the issues, challenges, and future research directions. The study aims to present an all-inclusive overview of deepfakes by providing insights into the deepfake creation techniques and the latest detection methods, facilitating the development of a robust and effective deepfake detection solution.

中文翻译：

---

Deepfake 视频：合成和检测技术——一项调查

深度学习已被用于计算机视觉中，以完成许多以前被认为过于复杂或资源密集而无法实现的任务。一个非凡的应用是创建 deepfakes。Deepfake 图像通过使用生成模型改变或操纵一个人的脸，以给出不同的表情或身份。应用于视频的 Deepfakes 可以通过某种方式改变面部表情，将不同的语音与最初给出的人关联起来。Deepfake 视频对法律、政治和社会系统构成严重威胁，因为它们会破坏一个人的完整性。正在设计研究解决方案来检测此类深度伪造内容，以保护隐私和打击虚假新闻。本研究详细介绍了现有的 deepfake 视频创建技术，并概述了公开可用的 deepfake 数据集。更重要的是，我们概述了 deepfake 检测方法，并讨论了问题、挑战和未来的研究方向。该研究旨在通过深入了解 deepfake 创建技术和最新检测方法，对 deepfake 进行全面概述，从而促进开发强大且有效的 deepfake 检测解决方案。