Advances in Deepfake Detection and Face Anti-Spoofing

The field of face recognition security is moving towards more sophisticated methods of detecting and preventing deepfakes and spoofing attacks. Researchers are exploring the potential of various deep learning models, including vision-language pretrained models and large language models, to improve the accuracy and generalizability of deepfake detection and face anti-spoofing systems. Notable advancements include the development of novel frameworks that leverage language-guided spoof cue map estimation and prompt-driven liveness feature disentanglement to enhance one-class face anti-spoofing models. Additionally, multi-modal learning capabilities and interpretable forged face detectors are being introduced to improve the generalization and explainability of deepfake detection. Some papers are particularly noteworthy, including the proposal of a spoof-aware one-class face anti-spoofing framework with language image pretraining, which consistently outperforms previous one-class face anti-spoofing methods. Another notable paper introduces a multi-modal face forgery detector that employs tailored face forgery prompt learning and incorporates a large language model to provide detailed textual explanations of its detection decisions, achieving state-of-the-art performance in detection and explanation generation tasks.

Sources

Face Spoofing Detection using Deep Learning

Unlocking the Hidden Potential of CLIP in Generalizable Deepfake Detection

SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining

Can Multi-modal (reasoning) LLMs work as deepfake detectors?

Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector

DF-Net: The Digital Forensics Network for Image Forgery Detection

DF2023: The Digital Forensics 2023 Dataset for Image Forgery Detection

Built with on top of