Ke Sun; Shen Chen; Taiping Yao; Xiaoshuai Sun; Shouhong Ding; Rongrong Ji

Link: https://link.springer.com/article/10.1007/s11263-024-02160-1

Published: 4 September 2024

Abstract

Recent research has made notable progress in addressing the growing challenge of face forgery detection. As face forgery techniques continue to advance, raising significant security concerns, a team of researchers has developed a promising new approach that aims to improve upon existing methods in both efficiency and effectiveness.

The study introduces the concept of Continual Face Forgery Detection (CFFD), an approach designed to learn from new forgery attacks while retaining the ability to detect previous ones. Central to this research is the Historical Distribution Preserving (HDP) framework, a system developed to maintain the distributions of historical faces.

By employing techniques such as universal adversarial perturbation (UAP) and knowledge distillation, the HDP framework works to simulate historical forgery distributions and preserve the distribution variation of real faces across different models. This approach seeks to address the resource-intensive nature of current methods, which often require substantial time and storage for fine-tuning on historical data.

To assess the effectiveness of their work, the research team has established a new benchmark for CFFD, including three evaluation protocols. Experiments conducted on these benchmarks suggest that the proposed method performs well compared to current alternatives.

This research contributes to the ongoing efforts in the field of digital security, offering a potentially more adaptable solution to the evolving challenge of face forgery detection. As digital threats continue to progress, this work represents another step in the collective effort to enhance our digital safeguards.

[Author Profile] Prof. Ji Rongrong is a Professor in School of Information, Xiamen University. He is also an Advisory Member for Artificial Intelligence Construction in the Electronic Information Education Committee of the National Ministry of Education. His research interests fall in the field of computer vision, multimedia analysis, and machine learning.