Research map / Federated learning and privacy / Privacy attacks and secure learning
Privacy-Preserving Face Recognition with Learnable Privacy Budgets in Frequency Domain
Jiazhen Ji, Huan Wang, Yuge Huang, Jiaxiang Wu, Xingkun Xu, Shouhong Ding, ShengChuan Zhang, Liujuan Cao, Rongrong Ji
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add frequency-domain transformation before noise (ablation) | DP face recognition · LFW | accuracy: 84.78 -> 99.48 | Improved |
| add frequency-domain transformation before noise (ablation) | DP face recognition · CFP-FP | accuracy: 63.87 -> 97.20 | Improved |
| add learnable privacy budgets instead of uniform budget (ablation) | DP face recognition · LFW | accuracy: 65.966 -> 99.48 | Improved |
| remove DC component (ablation) | BDCT baseline · LFW (no noise) | accuracy: 99.68 -> 99.48 | No clear effect |
| remove DC component (ablation) | BDCT baseline · CFP-FP (no noise) | accuracy: 98.3 -> 98.39 | No clear effect |
| remove DC component (ablation) | DP face recognition · IDCT visualization at budget 20 | face information leakage | Improved |
| apply method instead of PEEP | privacy-preserving face recognition · CFP-FP | accuracy: 74.47 -> 97.20 | Improved |
| apply method instead of InstaHide | privacy-preserving face recognition · IJB-C | TAR@FAR=1e-4: 69.02 -> 91.22 | Improved |
| apply method instead of Cloak | privacy-preserving face recognition · IJB-C | TAR@FAR=1e-4: 33.82 -> 91.22 | Improved |
| train on privacy-preserved VGGFace2 instead of original | ArcFace · LFW | accuracy: 99.60 -> 99.68 | No clear effect |
| train on privacy-preserved MS1M (budgets learned on VGGFace2) | ArcFace · CFP-FP | accuracy: 97.94 -> 96.94 | Hurt |
| lower the privacy budget (ablation) | DP face recognition · several face benchmarks | accuracy loss | Hurt |
Closest papers by meaning
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.89)
- Privacy-Preserving Face Recognition Using Trainable Feature Subtraction (CVPR 2024 · similarity 0.89)
- Privacy-Preserving Optics for Enhancing Protection in Face De-Identification (CVPR 2024 · similarity 0.86)
- How Private is Private? A Comparative Study for Face De-Identification (NeurIPS 2026 · similarity 0.85)
- Disguise without Disruption: Utility-Preserving Face De-identification (AAAI 2024 · similarity 0.85)
- CryptoFace: End-to-End Encrypted Face Recognition (CVPR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
