基於動態水印的人臉特徵深度偽造主動檢測方法
本文提出了一種基於動態水印的人臉特徵深度偽造主動檢測方法 (FaceProtect)，利用深度偽造過程中人臉特徵的變化作為一種新的檢測機制，有效提高了深度偽造檢測的準確性和泛化能力。
얼굴 특징 기반 동적 워터마크를 활용한 사전적 딥페이크 탐지 방법: FaceProtect
딥페이크 조작 전후의 얼굴 특징 변화를 활용하여 딥페이크를 탐지하는 새로운 사전적 접근 방식을 제안합니다.
Proactive Deepfake Detection Using Dynamic Watermarks Based on Facial Features
This research proposes a novel proactive approach to deepfake detection called FaceProtect, which leverages the unique characteristics of facial features to create dynamic watermarks, enhancing security and accuracy compared to existing methods.
時間步長生成：一種通用的合成 Deepfake 圖像檢測器
本文提出了一種名為時間步長生成（TSG）的新方法，用於檢測由深度學習模型生成的合成圖像（例如 Deepfake），該方法利用預先訓練的擴散模型網路作為特徵提取器，通過控制時間步長 t 來捕捉真實圖像和合成圖像之間的細微差異，並將這些特徵傳遞給分類器以進行檢測，實驗證明 TSG 在準確性和泛化性方面均優於現有方法。
Time Step Generating (TSG): A Novel Approach to Detecting Synthetic Images Generated by Diffusion Models
By leveraging the noise prediction capabilities of pre-trained diffusion models, Time Step Generating (TSG) offers a faster and more accurate method for detecting synthetic images compared to reconstruction-based approaches.
A Novel Meta-Learning Framework for Robust and Adaptable Deepfake Detection
This paper proposes a novel meta-learning framework that enhances the robustness and adaptability of deepfake detectors by addressing the challenges of generalization, adversarial robustness, and data drift.
Understanding Audiovisual Deepfake Detection: A Comprehensive Survey of Techniques, Challenges, and Datasets
Audiovisual deepfakes pose a growing threat, and this paper provides a comprehensive survey of detection techniques, emphasizing the importance of multimodal approaches for improved accuracy and robustness.
Human Perception of Audiovisual Deepfakes: Can We Detect Them?
While humans can detect audiovisual deepfakes slightly better than random chance, AI models significantly outperform them, highlighting the need for technological solutions to combat increasingly sophisticated deepfakes.
以品質為中心的通用 Deepfake 檢測框架
本文提出了一個以品質為中心的框架，用於解決 Deepfake 檢測中的泛化問題，通過評估偽造品質、增強低品質數據和採用課程學習策略來提高檢測模型的泛化能力。
A Quality-Centric Framework for Improving the Generalization of Deepfake Detection
This paper proposes a novel quality-centric framework to address the generalization issue in deepfake detection by leveraging forgery quality during training, enabling detectors to learn progressively from easy to hard deepfakes and avoid overfitting to low-quality artifacts.
