论文链接：DeepFakes and Beyond: A Survey ofFace Manipulation and Fake Detection
arXiv:2001.00179v1 [cs.CV] 1 Jan 2020
I. INTRODUCTION
II. TYPES OF FACIAL MANIPULATIONS
Facial manipulations can be categorised in four different groups regarding the level of manipulation [7]. Fig. 1 graphically summarises each facial manipulation group. We now describe each of them, from higher to lower level of manipulation:
- Face synthesis：this manipulation creates entire nonexistent faces, usually through powerful Generative Adversarial Networks (GANs) [18], e.g., through the recent StyleGAN approach proposed in [19]. （通过GAN直接生成现实中不存在的人脸）
- Face swap：this manipulation consists of replacing the face of one person with the face of another person. （用一个的脸去替换另一个的脸）。有两种方法可以实现：Two different approaches are usually considered: i) classical computer graphics-based techniques such as FaceSwap,and ii) novel deep learning techniques known as Deep-Fakes , e.g., the recent ZAO mobile application.
- Facial attributes: this manipulation consists of modifying some attributes of the face such as the colour of the hair or the skin, the gender, the age, adding glasses, etc [20].(修改人脸属性比如改变头发颜色，皮肤，年龄，添加眼镜)
- Facial expression: this manipulation consists of modifying the facial expression of the person, e.g., transferring the facial expression of one person to another person. One of the most popular techniques is Face2Face [22], working on real time. Recent approaches have shown its potential, generating high-quality videos of a person (Obama) changing what he is really saying in a target video [16].（此操作包括修改人的面部表情，例如，将一个人的面部表情转移到另一个人。Face2Face[22]是最流行的技术之一，用于实时工作。最近的方法已经显示了它的潜力，生成了一个人（奥巴马）改变他在目标视频中所说内容的高质量视频[16]。）
下面依次是每个方法的伪造方法和检测方法