Nowadays, Deepfake videos are widely spread over the Internet, which severely impairs the public trustworthiness and social security. Although more and more reliable detectors have recently sprung up for resisting against that new-emerging tampering technique, some challengeable issues still need to be addressed, such that most of Deepfake video detectors under the framework of the supervised mechanism require a large scale of samples with accurate labels for training. When the amount of the training samples with the true labels are not enough or the training data are maliciously poisoned by adversaries, the supervised classifier is probably not reliable for detection. To tackle that tough issue, it is proposed to design a fully unsupervised Deepfake detector. In particular, in the whole procedure of training or testing, we have no idea of any information about the true labels of samples. First, we novelly design a pseudo-label generator for labeling the training samples, where the traditional hand-crafted features are used to characterize both types of samples. Second, the training samples with the pseudo-labels are fed into the proposed enhanced contrastive learner, in which the discriminative features are further extracted and continually refined by iteration on the guidance of the contrastive loss. Last, relying on the inter-frame correlation, we complete the final binary classification between real and fake videos. A large scale of experimental results empirically verify the effectiveness of our proposed unsupervised Deepfake detector on the benchmark datasets including FF++, Celeb-DF, DFD, DFDC, and UADFV. Furthermore, our proposed well-performed detector is superior to the current unsupervised method, and comparable to the baseline supervised methods. More importantly, when facing the problem of the labeled data poisoned by malicious adversaries or insufficient data for training, our proposed unsupervised Deepfake detector performs its powerful superiority.

Recently, the rapid advancement of generative model has led to its exploitation by malicious actors who employ it to fabricate fake synthetic images. Meanwhile, the deceptive images are often disseminated on social network platforms, thereby undermining public trust. Although reliable forensic tools have emerged to detect generative fake images, the existing supervised detectors excessively rely on the correctly-labeled training samples, leading to overwhelming outsourcing annotation costs and the potential risk of suffering from label flipping attack. In light of the aforementioned limitations, we propose an unsupervised detector fighting against generative fake image. In particular, we assign the noisy labels to the training samples. Then dependent on the pre-clustered samples with noisy labels, the strategy of pre-training and re-training mechanism helps us train the feature extractor utilized to extract the discriminative feature. Last, the extracted feature guides us to respectively cluster both pristine and fake images; the fake images are effectively filtered by employing cosine similarity. Extensive experimental results highlight that our unsupervised detector rivals the baseline supervised methods; moreover, it has better capability of defending against label flipping attack.

The face manipulation technique such as Deepfake has been widely used to create realistic faces, which raises growing concerns in the community. Based on the correct labeled data, the current Deepfake detectors are mostly trained on the clean dataset, usually resulting in the reliable high detection accuracy. However, in the real-world scenario, labelers possibly mislabel the data or malicious attackers always intend to poison the training data with incorrect label, namely noisy label attack, leading to poor detection results. To overcome the tough issue, we propose a Deepfake detection framework fighting against noisy label attack. Specifically, a Negative Sample Generator (NSG) utilizes the possibly-poisoned samples to generate labelreliable negative samples through simulating blending artifacts caused by Deepfake. Next, a Noise-immune Contrastive Learner (NiCL) takes both positive and negative samples as training data, exploring blending artifacts and intrinsic forgery clues to filtrate the noisy samples out. Moreover, relying on label purification, the filtrated noisy samples are further purified, which then are fed back to the feature extractor for the following model training. Extensive experiments on the benchmark datasets demonstrate the superiority of our proposed Deepfake detector. In particular, when fighting against noisy label attack, the high performance of the proposed detector is remarkably better than its competitors.

With the continuous development of computer hardware equipment and deep learning technology, it is easier for people to swap faces in videos by currently-emerging multimedia tampering tools, such as the most popular deepfake. It would bring a series of new threats of security. Although many forensic researches have focused on this new type of manipulation and achieved high detection accuracy, most of which are based on supervised learning mechanism with requiring a large number of labeled samples for training. In this paper, we first develop a novel unsupervised detection manner for identifying deepfake videos. The main fundamental behind our proposed method is that the face region in the real video is taken by the camera while its counterpart in the deepfake video is usually generated by the computer; the provenance of two videos is totally different. Specifically, our method includes two clustering stages based on Photo-Response Non-Uniformity (PRNU) and noiseprint feature. Firstly, the PRNU fingerprint of each video frame is extracted, which is used to cluster the full-size identical source video (regardless of its real or fake). Secondly, we extract the noiseprint from the face region of the video, which is used to identify (re-cluster for the task of binary classification) the deepfake sample in each cluster. Numerical experiments verify our proposed unsupervised method performs very well on our own dataset and the benchmark FF++ dataset. More importantly, its performance rivals that of the supervised-based state-of-the-art detectors.