记一篇DeepFake主动防御方向的review

Contribution of the Submission

This paper is well-researched and meticulously written, offering a novel proactive defense strategy against DeepFake technology. By introducing intentional perturbations before media files are uploaded onto public networks, this method effectively renders DeepFAKE attacks ineffective. The proposed scheme achieves excellent results in experimental tests and demonstrates a robust defense mechanism while maintaining high levels of imperceptibility for any disturbances introduced.

Strength:

1. Discloses significant problems in prior DeepFake detection systems, particularly those involving passive detection mechanisms.
2. The proposed approach introduces a novel method for incorporating unique innovations into the HSV space.

Weakness:

There are certain challenges that must be overcome before a paper can be approved for publication. If these subsequent issues are properly resolved, this reviewer thinks that the fundamental contribution of this paper is crucial to advancing DeepFake defense.

1. 通过加入扰动手段实现主动防御的相关研究已有提及（见Ref.1），因此整体创新性不足。
2. 本文存在明显问题：缺乏对增强鲁棒性的解释。作者尚未进行HSV与RGB对比实验（如对比实验结果），无法证明通过加入HSV扰动比RGB扰动能获得更好的鲁棒性。
3. 实验数据集需要进一步优化。本文中所使用的Celeb-DF数据集主要用于评估检测算法的跨数据库性能。然而该数据集的伪造方法及数据类型单一，在现有研究基础上未能充分展现本文提出方法的优势及适用性；建议补充其他领域数据集上的实验结果。
4. 结论部分未对实验结果进行总结回顾，并对未来研究方向表述较为笼统。建议作者着重总结本文方法的关键实验结果，并补充相关思考内容。

Reference:

HUANG Q et al. (2021). Initiative Defense against Facial Manipulation. Proceedings of the AAAI Conference on Artificial Intelligence, 35(2):1619-1627.